Citigroup analyst Glen Yeung reiterates his Buy rating and $29 price target on Intel (INTC), citing channel checks that revealed strong demand for INTC's Atom processor:
"Field checks from Japan suggest Intel has ordered 20M Atom flip-chip packages for 3Q08 and 25M for 4Q08. This is orders of magnitude more than our modeled 2M and 2.5M units in that timeframe. These figures do sound aggressive to us (and some is likely inventory build) and so we believe a substantive haircut is appropriate. Nonetheless, even at half these levels, Atom is running well ahead of our expectations. We conclude that Intel's relatively aggressive 8.8% 3Q08 revenue growth guidance, while still prone to macro factors, looks more achievable in light of Atom strength".
I may have to give anecdotal evidence in support of Glen Yeung’s report. The lead time for ordering Atom processors from distributors have now gone up to 6 weeks! It's beginning to look like last year with Barcelona but only different. Atom demand has just gone through the roof as if every single technology firm has found some clever use for the low power microprocessor. Hopefully, Intel meets demand for the coming Christmas season. The concern about sales cannibalisation should be limited to computers below the performance range of Core 2 Duo, which as it stands today, equates to all of AMD’s SKUs.
8.07.2008
Strong Demand for Intel's Atom
Subscribe to:
Post Comments (Atom)
290 comments:
1 – 200 of 290 Newer› Newest»The concern about sales cannibalisation should be limited to computers below performance range of Core 2 Duo, which as it stands today, equates to all of AMD’s SKUs.
Okay, I must admit I exaggerated there a little bit…
I should have said Celeron.
is there anything wrong with the blogger.com? your previous post is missing some latest commments .. i thought it should be around 19x comments since 2 days ago and I saw Hyc's reply to my latest port too.
Yeah, the front page was showing 190 comments, but the link only showed 180 comments. Then a couple more of the old comments reappeared, and right now for me it shows 183, so 7 are still missing. Gotta love technology...
Hmmm, did an anonymous poster take issue with me regarding my prediction of INTC reaching the “high twenties”? I guess the glass is half full, after all. INTC’s moving average since the mortgage/bank debacle has only gone in one direction, up.
I still say when the dust settles, and when we get a turnaround, INTC will take off like a shot. Frankly, their business has never looked better.
I hate to use this expression, but the “tick tock” model seems to be working flawlessly.
While we cooking this stew, add this to the pot.
http://www.maximumpc.com/article/features/exclusive_we_build_first_nehalem_system_dont_tell_intel
Nehalem is the worst kept secret in the industry
AMD should be afraid, very afraid..
SPARKS
Sure the sales of atom may be running OK... but can it keep up with DTX, which clearly has taken off as predicted?
All kidding aside, Intel is going to need to beef up the performance a bit for the netbook/low end sector or sales will level off. And they seriously need to get to an SOC solution of something clearly better than 945. I think both of these things are on the roadmap but it is going to come down to how well Intel executes on the 2nd iteration of Atom, code name "molecule" TM.
Hmmm, did an anonymous poster take issue with me regarding my prediction of INTC reaching the “high twenties”
That would be me, but if I recall correctly I think I said 30's is too optimistic and even high 20's might be out of reach. As the stock is not even at 24, I'm not going to retract anything :) Of course I thought AMD would be putting up more of a fight by now as well.
Tech in general will likely be the first to move when things turnaround (and believe it or not probably the homebuilders too)
posts from the previous blog are back. there seem to be a blogger bug where the last 11 posts are no shown. So i had to make 12 "test" posts to make everyone elses appear.
hyc: I've seen plenty of posts in overclocking forums where SuperPi is provided as *one of* the results.
anonymous: Not to belabor the point, but there is a HUGE difference between someone using SuperPi to show the result of an overclock and the person using SuperPi to demonstrate stability (which is what you insinuated)
Uh, no. I didn't insinuate anything of the sort. I emphasized "one of" to reinforce the point that SuperPi was not being used all by itself. I was explicitly acknowledging that most people use multiple programs to demonstrate the stability of their OCs. You're claiming to be neutral and objective but you're going out of your way to mischaracterize my statements and intent.
ho ho: Wasnt' scientia supposed to be a programmer? If he is then he must be pretty bad one or retired for more than 15 years. Pretty much every programmer who knows anything knows that Java isn't interpreted language. He also seems to lack a lot of knowledge about JIT compiling and real-world experience on the field.
You're on shaky ground. A JIT compiler is just an interpreter. The native output of the interpreter is cached, but it's still not like the zero-overhead code of a truly native language. The JVM has orders of magnitude more overhead than natively compiled code, and if the application's data demands exceed the available cache, then the JIT code disappears and needs to be reinterpreted all over again the next time it needs to be executed. Any programmer who actually understands system architecture knows this, but only a small percentage of people in the world today who call themselves programmers actually understand computers...
Abinstein seems to be worried that Larrabee can run OS as fast as 1GHz Atom. Well, I've yet to see OS taking significant part of my CPU resources. It's the GUI rendering and user applications that does it. Also he seems to think that if you put OS on Larrabee it will be downgraded to single-core Pentium because old pentiums didn't support any of the Larrabee features. I wonder why is it so hard for him to understand that OS'es aren't programmed for specific CPUs and they use whatever features they expose like multiple cores and HT.
No. Most OS's *are* written for specific CPUs. Unix was fairly revolutionary in its time, because it was the first OS that wasn't written mainly in a processor-specific machine language, so it was actually portable to platforms other than what it was originally written for. Today Linux runs on a pretty wide variety of machines but guess what - that's because there's a whole slew of processor/architecture-specific source files off to the side, implementing a low-level set of features that the more generic OS code relies on. And whenever you decide you want to run the OS on a new CPU, hey, you get to write a new set of CPU-specific code.
"And whenever you decide you want to run the OS on a new CPU, hey, you get to write a new set of CPU-specific code."
So, how much changes have been made to, say, Windows 98, 2k or NT to make it work on today's CPUs? My point was if the CPU supports x86 it will be capable of running software (and OS) that was written for it. Only problem might be with CPU-specific features that need to be handled case-per-case basis.
This is completely off topic and not even new. But I find it interesting because it should be getting closer to implementation. The only thing I can find on timing is that Intel is planning to implement tri-gate transistors on either the 32 or 22nm node. So we may see an announcement on this as early as next year.
I believe this will be the next big HK/MG type breakthrough in power savings and performance.
The link at the bottom of the page leads to even more detailed links and a presentation for those interested in the more technical details.
More on topic, here we have a link that shows the battery life that the Menlow platform is capable of delivering. Moorestown is expected to at least double this performance.
Almost all the other reviews I've seen have been with the high power chipset designed for desktop use(name escapes me just now)and do not show the type of battery life this platform is capable of delivering.
Intheknow said...
I believe this [trigate transistors] will be the next big HK/MG type breakthrough in power savings and performance.
Very interesting - thanks for the link. I wonder how many extra processing steps this will require, and if it is only doable with immersion (hence 32 & 22nm nodes, which is where I believe Intel says they will use immersion).
Tri-Gate or what next year?
INTEL is sooooooooo far ahead of AMD right now I don't think they will need to pull the trigger on Tri-gate. Implementation of that with HighK metal gate at 32 nm would be adding risk on risk. SOmething INTEL doesn't need to stay competitive.
Today their design teams are on track and kicking AMD's A$$. Their process team has been kicking IBM and AMD's A$$ for 4 generations already.
I say they take it slow and enjoy the fruits of their labor. They will be first to 32nm and first to 22nm and by then AMD will be BK for sure.
Anonymous Anonymous said...
...
INTEL is sooooooooo far ahead of AMD right now I don't think they will need to pull the trigger on Tri-gate. Implementation of that with HighK metal gate at 32 nm would be adding risk on risk. SOmething INTEL doesn't need to stay competitive.
...
I say they take it slow and enjoy the fruits of their labor. They will be first to 32nm and first to 22nm and by then AMD will be BK for sure.
I guess you miss the link i provided in the comment of the earlier post.
http://www.hardspell.com/doc/hard/79579.htm
AMD's Deneb looks competitive in the early benchmark, at least compared to YFD? (assuming this allow them have higher clock part)
Pointer
AMD's Deneb looks competitive in the early benchmark
It looks like Deneb will be competitive with Kentsfield though probably not caught up with Yorkfield. However, it also appears that Nehalem will be out before Deneb, which will ensure that Deneb will be priced like the current Phenoms at the highest.
The low ASP trend will continue for the forseeable future and AMD will be unable to raise their breakeven point above the current target of $1.5 billion per quarter for quite a while. This means insufficient funding for the next process node, meaning that 45-nm is likely to be the final in-house fabrication technology for AMD.
Bk I said, as each new AMD processor comes out INTEL has the response on deck.
Like I said with design in order, far superior processing technology, huge econcomies of scale of their 65nm and 45nm factories AMD can't hope to compete AND MAKE MONEY.
Sure they can be competitive, but that ISN'T ENOUGH! They NEED to MAKE MONEY, and they WON'T
BK like sharikou says, LOL
hyc said: If it's only stable enough to boot and get SuperPi running, but no other apps work, then it's pointless, let alone stable...
These are your exact words... to a normal reader, you make it sound as if some people are using a boot and superpi to claim stability (or believe this to be the case). If noone was, then why throw in the "let alone stable" part at the end? That makes it sound as if people are using these 2 things for a stability check (or believe it to be sufficient). That is why I thought/said you were insinuating it.
Please 'characterize' your quote up top - you said boot and superpi, and then felt the need to say "let alone stable" - what exactly did you mean by this? Where was anyone saying that boot and superpi was enough for stability? And if that wasn't the case (and you weren't trying to insinuate that people were making this claim), why did you feel the need to go out of your way and state 'let alone stable' at the end? Why not just end your statement at 'pointless'?
Words do matter - we are not mind readers. Either you are waffling on your argument or you chose your words poorly (which is fine, we all do this). But don't say I'm mischaracterizing you - I'm only trying to understand the statements you make from the words you choose.
Where was anyone saying that boot and superpi was enough for stability?
Not to put words in hyc's mouth but I assumed he referred to SuperPi because that was the program under discussion at the time -- a sort of [insert program here] type deal. For example, I think over at Scientia's blog, Prime95 is supposedly the 'one true overclock benchmark (tm).
In The Know –
As I recall GURU was way on top of this tri-gate technology last year.
In fact, if memory serves, he said something to the effect that ‘if you really want to show me something impressive it would be tri gate transistors’. Not an exact quote but the essence is there.
I believe, at the time we were discussing the merits of EUV process technology. (I don’t think he was too impressed with EUV at the time.) For myself, I’ve been following the technology, and have done some searches. INTC looks like it has a handle on the process.
Nothing gets by GURU. Perhaps, with this discussion, we can flush him out for a review and a comment.
That said, with the way their executing, nothing coming from INTC surprises me these days. Honestly, in retrospect, I think the P4, shall we say misstep (?), has done more for INTC than anyone could have possibly envisioned 2 years ago.
SPARKS
Anonymous said...
hyc said: If it's only stable enough to boot and get SuperPi running, but no other apps work, then it's pointless, let alone stable...
These are your exact words... to a normal reader, you make it sound as if some people are using a boot and superpi to claim stability (or believe this to be the case). If noone was, then why throw in the "let alone stable" part at the end? That makes it sound as if people are using these 2 things for a stability check (or believe it to be sufficient). That is why I thought/said you were insinuating it.
Please 'characterize' your quote up top - you said boot and superpi, and then felt the need to say "let alone stable" - what exactly did you mean by this?
As I said in a response to pointer, the choice of SuperPi here was just an example, since we were discussing that and AMDzone at the time. I was attempting to clarify a position that was being attributed to the AMDzone regulars, but which I don't think had been represented accurately here. I didn't say "my point is" - I said "I think the point is" - i.e., the point the AMDzone folks were on about.
I agreed that what's acceptable to a particular user is different from the requirements that a CPU manufacturer expects. I agreed that in reality, few people (if any) are claiming OC stability with just a SuperPi run. I also further stated that I personally didn't find either SuperPi or Prime95 very relevant for my own computing, and would prefer a pure-integer benchmark.
If any of that wasn't clear, I apologize for my poor choice of words.
I don’t think he was too impressed with EUV at the time.
EUV is like immersion litho - call me when it can do something existing technologies cannot. I saw an AMD PRODUCT presentation (it was either Deneb or Shanghai) stating 45nm process, immersion lithography.
To this I say who cares what litho is used on 45nm? Does immersion make the chip any faster than alternative technologies? NO! Does it make it perform better? NO! Why point it out and make it sound like it is doing something special for the products. Why not say 45nm, 0.5um ILD3 layer technology? or 45nm, 50% unlanded contact technology?
Why does AMD advertise the lithography? It is either an inferiority complex ("look we can do something Intel hasn't implemented yet") or an attempt to BS the media and fanboys into thinking immersion confers some specific benefit to the product. It is simply an alternative way to print features and provided you can print the same features, it really doesn't matter.
The joke of it all is the more impressive feat is the ability to extend existing technology further and getting the same that the new technology enables (and doing it a year earlier). But what you will NEVER see from AMD is an actual specific benefit from a process perspective attributed to immersion litho - they will talk about it, couple it with some other changes and give an overall benefit which has nothing to do with immersion litho. (this is the standard shell or 3 card monty game)
EUV is much the same - if dry litho or immersion litho is extendable, then why care about the latest "blue crsytals". If it enables patterns smaller than current litho, then it is a whole new story. If Intel say, our new 45nm technology with EUV lithoography... I would say who cares... why would you do this if there are other approaches that can do the same thing?
Right now most indications are that double exposure immersion 193nm litho will work for 32nm, and I wouldn't put money against someone being able to figure out how to extend it to 22nm (whether it be an alternative medium to water for the immersion fluid or some other improvement). EUV still has a bunch of major technical hurdles and is still relegated to the next big thing (which it has been now for several technology nodes)
What makes this clear (in my view)is that Intel is still talking about it. Remember highK/metal gate? How much talk was there right before Intel implemented it? If they are talking about it is likely to keep suppliers and reearchers on board and behind it. If it was about to be implemented you would likely see (hear) silence.
"I also further stated that I personally didn't find either SuperPi or Prime95 very relevant for my own computing, and would prefer a pure-integer benchmark."
If you believed this then why not say:
If it's only stable enough to boot and get SuperPi running, but no other apps work, then it's pointless.
instead of this:
If it's only stable enough to boot and get SuperPi running, but no other apps work, then it's pointless, yet alone stable...
So I'll ask again... what was the point of saying "yet alone stable"? Why add that at the end?
Cannot you not see how this might be interpreted that you were inferring that some believe this constitutes stability, or why would you feel the need to say it?
And in the end you still refuse to let it go with your:
I agreed that in reality, few people (if any) are claiming OC stability with just a SuperPi run.
What is is few or none? Why even say few? Again you are trying to imply a few, but hedging on none. Do you have a link to people who claim this is stability? Who specifically on this board are you agreeing with that said a few people believe this?
And on a completely different note, back when I bought my Asus M6Ne laptop in 2004, I figured out how to overclock it a bit as well. First just playing with the memory timings
http://forum.notebookreview.com/archive/index.php/t-4392.html
later figured out how to adjust the 100MHz reference clock. Using DDR400 SODIMMs instead of DDR333, could bump the clock to 120MHz yielding 2.4GHz processor clock and DDR400 (effective) memory clock. I'm still using this laptop even now, but these days I'm more interested in undervolting and underclocking to extend the battery life.
(By the way, yeah, it would boot Windows and run SuperPi just fine at 2.4GHz, but would crash when I tried to compile code, so I eventually had to drop the multiplier back from x20 to x19 for 2.28GHz stable... Somewhere in this thread http://forum.notebookreview.com/showthread.php?t=41097&page=13 I posted my SuperPi times.)
Do you have a link to people who claim this is stability?
No, I don't. I haven't really paid any attention to the overclocking scene since 2004. Whatever I remember reading from back then is long since obsolete.
Ok, let’s put this speculative SuperPi speculative stuff to rest.
QX9770 1.41V
9X Multiplier @ 450FSB
DDR3 1800MHz @2.10 V
4.06 GHz
I, ME, PERSONALLY ran Super Pi, today. August 8 2008, 9:00 PM, EST
These runs were with Task Manager right on top, GOT IT?
First SUPER PI DOES NOT LOAD ALL 4 CORES 100%, HELLO!!!!
Second, TOTAL CPU USAGE for a 1M run was only 28%, imagine that!!!!
I got 10 seconds, WHO GAF!
Soooooooooo-------------------------
I ran 8M it finished in 2m 24s, who cares.
Core 0 was pissing along between 20 and 30%
Core 1 was working between 50 and 90%
Core 2 ditto as core 1
Core 3 was doing squat between 5 and 25%
Fact-SUPER PI DOES NOT LOAD THE CPU ALL THE TIME THE WAY PRIME95 DOES.
Soooooooooo----------------------------
I did the 8M run again, this time, while running Norton AV AND Diskeeper Defrag.
The cores had the same load ratio each, but TOTAL CPU usage went from 35 to 50%.
The difference was ONLY 2 seconds @ 2M 26s.
Fact-Super Pi does not “heat” or load the CPU ANYTHING like that abomination Prime95. Further, there was no appreciable affect in time.
The machine didn’t miss a beat, it didn’t fart, belch, crash, burn, puke nothin’, zip, nada!
So I ask you, what the hell is it good for?
SPARKS
“If it was about to be implemented you would likely see (hear) silence.”
Perhaps that why we are hearing so little of the energy efficient, low threshold voltage tri-gate transistors? After all, you have only mentioned them one time since.
Therefore, they must be close. These are dated 1H 2006.
http://www.intel.com/technology/silicon/tri-gate-demonstrated.htm
http://www.physorg.com/news69432815.html
http://www.intel.com/pressroom/archive/releases/20060612corp.htm
http://news.cnet.com/Intel-puts-Tri-Gate-transistor-on-fast-track/2100-1006_3-1015424.html
Then nothing.
But, they are pretty bastards. Hey, nothing like thinking/switching in 3D.
http://reviews.cnet.com/i/blog2/20060612/Intel_tri-gate-1150128337791-200_150.jpg
SPARKS
"No, I don't. I haven't really paid any attention to the overclocking scene since 2004. Whatever I remember reading from back then is long since obsolete."
Great, so your comments refer to something you remember reading prior to 2004 and something that you really haven't paid attention to for 4 years... thanks for discussing something that you now consider obsolete and cannot support. (Let me know, if I'm mischaracterizing)
To this I say who cares what litho is used on 45nm? Does immersion make the chip any faster than alternative technologies? NO! Does it make it perform better? NO! Why point it out and make it sound like it is doing something special for the products.
I remember reading something about Intel's DFM process a while back (and think I posted a link on this site). They had an image that showed some unusual restrictions on the patterning of Intel's product. The article attributed these restrictions on the use of double patterned dry litho.
Now for all I know, the author could have been all wet and it could have been a restriction to accommodate polish, or back end interconnects or any of a dozen other things. But if the speculation was correct, then immersion litho does bring something to the table. In that case, it would lift design restrictions and possibly make room for some additional functionality.
And then it would be a competitive advantage.
I'm not saying it is, or it isn't. I'm just saying that dismissing it out of hand may be taking too narrow of a viewpoint.
Yeah, I know. Don't feed the troll. But I just can't resist the temptation to wind him up and watch him prance about.
INTEL is sooooooooo far ahead of AMD right now I don't think they will need to pull the trigger on Tri-gate. Implementation of that with HighK metal gate at 32 nm would be adding risk on risk. SOmething INTEL doesn't need to stay competitive.
What you are displaying here is the kind of overweening pride that lead to P4. Now you want Intel to repeat that mistake? Sit back on their laurels and watch the competition try and catch up? Perhaps you forgot that "only the paranoid survive."
And you are way too focused on / obsessed with AMD. If Intel is going to move down the applications stack, they have got to make serious improvements on the perfomance/power curve. To get into smart phones and other CE products they need much longer battery life. So they have got to drop the power regardless of what AMD and IBM can or can't pull out of their hat.
Oh, and maybe you should read the link. Intel's paper on tri-gate from 2 years ago already incorporated HK/MG. So the only extra risk here is in your imagination.
Sparks said...
Perhaps that why we are hearing so little of the energy efficient, low threshold voltage tri-gate transistors? After all, you have only mentioned them one time since.
Which, when combined with the approaching advent of 32nm in a year's time is why I brought up a 2 year old subject. With all the silence, I gotta wonder if this one isn't getting real close to being rolled off the production line.
ITK- Well, that just about covers that.
Man, we had a lot of rain on the East Coast today.
SPARKS
"Sit back on their laurels and watch the competition try and catch up?"
Not with Big Paulie ya don't.
SPARKS
sparks: "So I ask you, what the hell is it good for?"
I don't think anyone uses SuperPi for anything other than a quick "check out this score" snapshot when they are overclocking. In one of the forums (don't recall which one... extreme tech maybe) when someone is showing his overlocking, they will have a screenshot of a desktop showing a CPUID program (so you can see the CPU frequency), and then a couple of applications like SuperPi and Cinema 4D. It's really a quick and dirty look at the overclock.
I don't think the forumers at AMD Zone are impressed with SuperPi either. Scientia feels that Prime95 (with specific parameters) is what should be used to test OC stability, but I don't think anyone really sees SuperPi as a true test of anything.
It is sort of a "much ado over nothing" situation, really.
axel said...
The low ASP trend will continue for the forseeable future and AMD will be unable to raise their breakeven point above the current target of $1.5 billion per quarter for quite a while.
But Dementia over on AMDZone is predicting AMD will return to profitability in Q4! Of course, he forgot to say which year :) but presumably he meant this year. Guess he is banking on DTX - er, make that Puma - and Deneb and 790G and I dunno what else.
OTOH, he started a thread there where he basically now agrees AMD had a problem at 65nm (after months of flaming those who said so). Apparently he changed his mind based on - get this - papers published about Intel's difficulties transitioning to 65nm and what they had to do to get there, and extrapolating that experience to AMD. Of course, at this point I don't trust any of his 'informed' guesses, but it does hang together logically at first glance.
Anyway, Dementia thinks AMD will experience a much smoother transition to 45nm.. Smart money says - we'll see.
Anyway, Dementia thinks AMD will experience a much smoother transition to 45nm.. Smart money says - we'll see.
Because it only gets easier as the features get smaller, of course.
a nonny moose
But Dementia over on AMDZone is predicting AMD will return to profitability in Q4!
Well, AMD actually do have a good chance of making an operating profit in Q4. But the main reason for this is not revenue growth, it's the great reduction in quarterly CAPEX and R&D. They are clearly curtailing their investments into future fabrication technology.
Well, AMD actually do have a good chance of making an operating profit in Q4.
Have to agree with Axel as well, Though there are a few problems even if this is achieved:
1) Operating profit does not equal net profit, so even if AMD achieves operating profit they will in all likelihood still have an unprofitable quarter. AMD has already been out ahead on the spin on this so this will get completely lost in the press.
2) Along Axel's line of thought - you can cut your way to (operating) profitability but it is not sustainable and profitability does not necessarily equal GROWTH. If you achieve profitability by cutting back capital expenditures, what happens when you need to eventually make those spends in the future? Q1 and Q2 are typically seasonally down quarters so I'm not seeing where the cash flow is.
Not too many companies cut their way (via headcount and capital spending) to growth. Granted AMD's position has forced them into this, but again it is not sustainable so it will be interesting to see what AMD has planned beyond cutting costs.
“That would be me, but if I recall correctly I think I said 30's is too optimistic and even high 20's might be out of reach. As the stock is not even at 24, I'm not going to retract anything :)”
That’s precisely what you said. Retraction, you say? This is not at all necessary or expected. No one on this site ever backs down.
They just get proven wrong. :)
By the way, the stock IS over 24. It only took one day, Friday; it went up another 56 cents. This is very close to the mid twenties, yes?
Hmmm, here’s to some back to school specials, peppered with a dash of Nehalem, bake ~1200 C to a full ramp at 45, serve warm at Thanksgiving.
Just to be clear, “30’s” is a penny over 30 dollars, and “high” twenties is a penny over 28, right?
Cheers, and a Merry Christmas, to all!
SPARKS
Oh, and maybe you should read the link. Intel's paper on tri-gate from 2 years ago already incorporated HK/MG. So the only extra risk here is in your imagination.
InTheKnow - while I generally agree with your analysis, perhaps YOU TOO should read Intel's paper...
If you look it over carefully, you will see Intel clearly uses a GATE FIRST flow - they talked about a mid-gap metal via PVD, followed by Poly Si CVD for the gate and also mentioned characteristics on the gate etch (which wouldn't be mentioned if you were doing a gate last flow). In other words the gate (which probably was a simple TiN/Poly Si cap) and the integration flow in this paper are not the same as Intel's 45nm process which is fairly common for research work when trying to prove out feasibility.
While these are high performance devices, I'm not sure if they would be production worthy with the mid-gap solution. As many know, Intel uses a dual metal, band edge matched solution on 45nm as a mid-gap metal means a performance hit.
So why is this important? If Intel sticks with a dual metal, gate last solution you now have a different geometry you are filling and this could lead to issues. With a planar transistor (45nm process), you are simply filling a trench with your work function metals and filling the rest with your contact metal. With tri-gate you would have a more complex geometry you are filling - this could lead to either coverage issues with the workfuntion metals or a higher chance of voids with the fill metal.
It's hard to say how big an issue this might be, but the paper wouldn't (and doesn't) address this. Now Intel could be planning to switch to a gate first flow when implementing tri-gate, but this would probably make a 32nm implementation less likely as Intel will rarely use a large integration change (45nm gate last flow) for just a one generation solution.
In any event the link you provided, while informative, doesn't address a potential combined tri-gate / gate last solution. It is hard to say how big a risk it is, but it shouldn't be simply dismissed (nor does Intel's published works address it).
Just to be clear, “30’s” is a penny over 30 dollars, and “high” twenties is a penny over 28, right?
Yes to 30's and I'll give you anything over 27.50!!!! :)
Keep in mind 30's would mean another >20% move and the macroeconomic conditions will matter more than AMD or Intel execution in achieving this.
Don't get me wrong, I have some Intel stock and would love to see it get into the 30's but I'm a pessimist by nature (and may sell if it gets in the 26ish range and switch over to RIMM)
LOL if it is published its a done deal.
I think you missed the biggest challenge with Tri-gate you process expert god.
By the way process technology isn't like architecture and design...
Process technology is the building block for design.
Once ahead it provide the basis to implement your design. Once ahead it enables lots of things. What it can't do is fix a bad design or architecture decision. The discussion about Pentium IV has nothing to do with process or manufacturing lead. INTEL had it then and has a bigger lead now. With that lead it makes little sense to leap to a radically new and more complex process. Having process leadership doesn't gurantee anything. INTEL senior managers need to make sure that process superiority and manufacturing gets used for efficent designs, not like in the Pentium IV days when they made space heaters.
From what I read INTEL lead looks good with Nehalem, Penrym, and Silverthorn. AMD's clock has run out; tick tock tick tock. As long as INTEL's next Tick and Tock keep the pace AMD is finished.
“LOL if it is published its a done deal.”
Ah, I don’t think so. I’m no GURU, by any stretch.
“For most people in the world who lack experience in fab processes, subtle complexities get lost in translation and details are distorted or lost. The truth about nanometer-era fab processes is that they are all tough to develop with inherent integration trade-offs. With Intel and IBM now going down divergent paths, it’s truly difficult to assess a fab technology “leader” in anything but hype”
The key phases here are “subtle complexities” and “inherent integration trade-offs”.
I think if you really have an inside track to process development, you can speak from experience on what’s beyond published knowledge. I certainly can’t add 2 plus two here.
What make you think the others here, who cook this stuff for a living, can’t?
Can you?
http://www.pennwellblogs.com/sst/eds_threads/2007/05/070525-intel-ibm-fab-hype-war-and.php
SPARKS
Sparks, it was tongue in cheek comment. Publishing doesn't mean a thing. IBM published a ton of neat papers on 3-d devices, NONE came to be seen in a chip. One device published says nothing about what it takes to manufacturing hundreds of millions of chip with perhaps a billion transistors ALL working.
As to your link, IBM is well known for lots of press. They were once unquestioned leaders in silicon but no longer.
Their SILK was a disaster
SOI gets lot of press, but INTEL and TSMC all are very competitive without this more expensive process.
They are a generation late on HighK. They were totally surprised by INTEL and at the last momenet tried to beat INTEL with a weekend press anoucement. INTEL is in production with the biggest change in silicon manufacturing, IBM has NOTHING coming till 32nm at the earliest.
The clock has run out on AMD and the consortium. You band togather when you can't afford to do it along. Consortiums are like the United Nations compromises and tradeoffs that result in a medicore process.
Lex... umm... I'm sorry "anonymous", a few things you might want to be aware of.
They were totally surprised by INTEL and at the last momenet tried to beat INTEL with a weekend press anoucement.
Actually both were aware of what each other was doing and both were trying to beat the other to the punch on the announcement - though it really didn't matter because Intel was announcing "going into production" and IBM was announcing "we're pretty sure we can go into production at some point", so in the end the timing of the announcement was meaningless and the "WHEN" was the more meaningful part. Announcements give you a few days or PR and make senior managers feel good (especially if they are first), the substance of the announcement is what matters in the end.
This is much like your comments, lots of flash and occasionally you'll try to sprinkle in some fancy jargon to make it almost sound like you know what you are talking about but if one reads carefully, there is very little actual substance or any actual original thought.
Reminds me of the Michael Bolton look-a-like in Goodwill Hunting in the scene in the Haa-vaad Baaa where he is simply quoting other people to make himself look good and can't seem to put together an original thought. How do you like them Apples!
I think you missed the biggest challenge with Tri-gate you process expert god.
Let me guess, it's confidential... I didn't realize the manufacturing folks at Intel were exposed to much sensitive material (maybe a D1C/D1D sustaining person with a resentment of the process development folks or AZ/NM with just a distaste for the Portland folks?) I can help you sound out the big words if needed, and no I don't work at Intel so you'll have to be careful that you are not using Intel specific terms.
And FYI, I wasn't attempting to describe the biggest challenges to tri-gate - I was trying to point out (and unlike you do it in a semi-constructive way) that intheknow wasn't quite interpreting the Intel paper correctly and to present a case where the current Intel highK solution combined with tri-gate could introduce some additional challenges unique to the combination of the two (I notice you did not disagree with any of it).
So do us a favor and post something constructive with some actual thoughts - like your comment comparing process consortiums to the UN, that was actually something which seemed to be reasonably thought out and useful, and not just your typical flaming.
Folks by now can clearly see Mr Tick Tock with a lot of 'wrong' 'you missed the boat' comments, but no actual substance, original thoughts or information to support what he is saying.
Sorry, it’s hard to tell, it’s those “subtle complexities” between you guys that I miss.
In any event ITK, managed to flush out GURU, and we’ve picked up some interesting links, as well. Well, at least I have.
All this discussion make me feel like a kid again, learning how the ‘HeathKit’ servo I just soldered together works. Hell, I’m still coming to terms with the differences/similarities of ALD, CVD, and PVD.
http://www.solid-state.com/display_article/205422/5/none/none/Dept/High-k-gate-deposition:-ALD-or-CVD
This article may be old as the hills, but there is a wealth of information for the newbe.
In any event, I’m looking forward to INTC’s successful implementation of Tri Gate just to see if they sprayed the gates on the top or on the bottom. I don’t think we will ever know how the dealt will the critical temperatures.
Finally, I’m looking forward to the new technology for a future build, of course, so EVERYONE can LUST on my newest killer rig
After all, that’s why you guy’s are in business. And, I hope the boys at INTC never forget that-----------, AGAIN!
SPARKS
Being a nice lazy Sunday, with coffee in hand and QX9770 seductively clocking along, I ran across this ‘EE Times’ article. It is titled,
“Intel Core i7 CPUs sport energy saving feature
Company's first 45nm desktop chips will be focus of IDF”
I rolled over and asked my lovely QX9770, “My love, aren’t YOU a 45nM chip?”
She replied, “Yes, I am my love, why don’t you take me to 4.26 GHz, and I’ll show you how much I love you. I said, “No, I love just the way you are.”
I continued to read on, and then I read this:
“The chips are Intel's first to integrate a memory controller and high-speed, cache-coherent interconnect, following in the footsteps of archrival Advanced Micro Devices.”
I rolled over again with a hug asked, “Sweetheart, didn’t you have a cousin ‘Timna' who tried to make it big about 10 years ago with her IMC, but she didn’t get any support from the family?”
“Yes”, she replied, “Don’t listen to those terrible people, they don’t know what their talking about. Timna even had an integrated graphics solution, too.”
Then she gave me a wink, and I had to do it, I did the nasty. I took her to 4.26 Gig. We went to “The Harbor” in “Crysis” and slaughtered some North Koreans, hand to hand. Ah, if they were only dweebie journalists or market anal-ists.
What a doll, she just wore me out, as usual.
SPARKS
http://www.eetimes.com/news/latest/showArticle.jhtml;jsessionid=TSZOG1PBB53KYQSNDLPCKHSCJUNN2JVN?articleID=210000564
My My My there is a big arrogrant anonymous poster that picks on little old Lex, can't find satisfaction in your hard job?
Got to put down people in other organizations, let me guess you are one overworked and underappreciated person in development maybe?
I seem to be in error. AMD’s bond rating is not BBB-.
Fitch has affirmed it’s B- rating, issuer default, also known as Junk.
http://news.moneycentral.msn.com/ticker/article.aspx?Feed=AP&Date=20080808&ID=9007658&Symbol=AMD
This is not good.
http://upload.wikimedia.org/wikipedia/ commons/c/c3/Main_Credit_Ratings.png
SPARKS
"My My My there is a big arrogrant anonymous poster that picks on little old Lex"
You've got to be kidding - go back and look through the old posts of this Mr Tick Tock guy. Both on this board and Scientia's (under the name lex) before he got banned there.
I have no problem with strong opinions, but this guy has been vulgar, disrespectful and often times wrong or misleading on the facts.
Give me the other anonymous poster any day of the week. He at least offers up some real information, tries to explain things and seems to only take issue with one specific person and in that case he gets a pass from me. It's not like he's picked on anyone else.
And I've got to wonder if the person above Sparks is just Mr Tick Tock, with the sig line missing trying to drum up some sympathy. Before anyone feels sympathy let's pull out some of Tick Tock's gems:
"As to the AMD cum lappers should focus on executing versus worrying about the competition. " June 14th
"Sorry meet me over beers, can;t talk about that stuff on the net." June 15
"Toxic chemicals everywhere when you lose power, not likely except maybe in some 3rd world country." June 15
"Most AMD fanboys really don’t realize this but blindly lap at competition even from an inferior and incompenent little dick is better then letting the marketing efficiencys play out." June 24
"The only logical conclusion, this guy is one dumb fuck!" June 25
"Now some retard talked about PhDs and fancy machines to figure out gates" June 26
"LOL you an idiot or what" June 29
"Dude are you a designer or a retard?" June 29
"You are either a silly arrogrant INTEL TD guy who doesn't toe the party line or a jealous INTEL wanna be." July 5
"That makes you a sad loser does it not." July 6
"You arm chair QB got no f**cking clue." May 14
"Anyone who thinks you aren't limited by your weakest link is a retard." May 14
"In the end I’m surprised the AMD cum lappers aren’t celebrating, but in the end its clear why. " May 30
And From Scientia's blog, under the name Lex:
"Superior process with competent design is a much better situation then great design on a inferior and lagging technology node." Does this theme sound familiar at all?
"Tick Tock Tick Tock by the time they got that squared away on 65nm..." Sound familiar?
"YOu talk a good story about architecture but when it comes to technology you are an idiot who doesn't know the first thing about process."
"It amazes me that the cheerleaders don't really understand why INTEL is winning or the fact that INTEL will win the war and even more amazing how the fanbois can't appreciate why AMD will lose the war regardless of their one trick poney and FUD they continue to spout
Tick Tock Tick Tock"
Tag line anyone?
"Tick tock tick tock AMD is seing its limted life down to its last tock."
"It is funny to keep yanking you limp AMD fanbois. AMD is finished, don't let a slight uptick or another disastrious multi-hundred million loss with a good Hector/Dirk talk about bright future get you a booner. You can rub it as hard as you like it won't change the fact you are still going to go limp. AMD got no mojo plain and simple."
By the way "Lex" did list Arizona in his profile - that does help explain a lot.
I'm sorry, who was the arrogant one again?
That is one unhealthy obsession with wanking off and lapping cum. Ewww...
I think someone has an unhealthy obsession with some dude name Lex and some other guy that likes clocks, NO?
Mighty impressive catalog of quotes
I think someone has an unhealthy obsession with some dude name Lex and some other guy that likes clocks, NO?
I think it's fairly obvious they are the same person, kind of curious why Mr.Tock doesn't use his sign-in; as there is no 'other clock guy'. Good find on the AZ information as well.
Hey guys! Come on. Can't we talk computers, heh?
Jeese, I just bought a copy of 'CPU' Mag, is any body interested in a report, or are we going to comment about rudimentary biological secretions? Eeeww.
INTC is up 15 cents on the day.
Come on, anything! I'm jones'n.
SPARKS
'Core i7'? I have a lot of respect for Intel's marketing (people laughed at "core" and "atom")... but i7? WTF?
I can see it in the stores now:
"You don't want the Core 2, you want the Core i7."
Yes Core 3 would have been boring but it reinforces the brand better. It's almost as if the marketing team felt the need to justify its existence. KISS (keep it simple, stupid)
To be honest, I'm just as perplexed with Core i7 as the next guy. I can't seem to figure out how that enhances product recognition or improves the brand identity. I've discussed it with other folks and they're also confused. If Intel employees can't figure it out, I don't know how Joe Sixpack will.
I don't have any detailed info beyond what has already been reported, but I get the feeling that the Core i7 will be Bloomfield ONLY. The Lynnfield and Havendale variants will receive their own xM moniker which will effectively serve the purpose of E7XXX, E8XXX and Q9XXX models and remove redundancy's like the Core 2 Duo presents when explained to people.
Also, it appears that i7 has no hidden or further meaning (like intel 7, or 7th intel architecture etc...) the number is meaningless. It will be used as an arbitrary value to measure against other SKU's on a relative scale like xM vs yN vs zO kind of thing. Whether or not this pans out and is more clear down the road is anyone's guess.
i7 WTF
You know engineers weren't part of the decision process.
It was probably the result of a bunch of marketing monkeys, customer surveys and some slick powerpoint foils that sold Paul.
Where are the days where the label told you something about the product?
But again, INTEL sells computers and INTEL inside to many people who could care little and probably comprehend less about MegHz, Quickpath, 4 cores, 7 thread etc. etc. They only want to know, is it fast enough for xyz, will it run this game, or is it the best deal.
And thus the marketing monkey with no clue to technology becomes king in the corner office.
At least they gave some lineage as the other poster noted to "7" for 7th generation?
You guys like the latest news on Business week online? Its about how AMD might really go Ass light and sell their fabs. I wonder if they do that do they have to share their non-existent profits with the owner. Clearly the owner of the fabs will want some serious bucks to run it, and also even more serious bucks to upgrade to 32nm. And if they want to run a complicated say 32nm HighK, airgap process who is going to pay. TSMC isn't into glory, they want to attract multiple customers to make money. Ass light is the end of AMD being a leading edge competitive CPU company. They can only survive ASS light if they choose to not compete directly with INTEL. Superior technology is a critical requirement and going ass light will foever leave them even further behind.
The clock has run out. Ruiz finished them when he could have saved them...
Oh well debating AMD and INTEL is very boring these days.
I want to see more quotes that from lex from years back, can we please?
"32nm HighK, airgap process who is going to pay."
Now we’re talking! Nano Vacuum tubes!
Er= Air=1.0054
Er= Vacuum=1(by standard definition)
NICE!
SPARKS
BTW:
SiO2=Er=3.7
SPARKS
"I want to see more quotes that from lex from years back, can we please?"
You don't store what you type? Do you have a split personality or you just trying to be 'tricky'? What's the weather like in Arizona?
And by the way those quotes were from ~6-7 months ago, then Scientia started editing them, and then through a remarkable coincidence "the clock has run out" guy showed up here. It is laughable for you to try to pretend... the fundamental problem is that while you post anonymously, you can't take BEING anonymous - thus the real 'catchy' catchphrase.
Robo - can you track this guy's IP and expose him for who he is?
Sparks, I can't believe you sold out. I mean, going with the blue vanilla logo. And here I thought you were into the high end. Don't you know the extreme edition is identified by the grey scale logo? ;-P
f you look it over carefully, you will see Intel clearly uses a GATE FIRST flow
You are absolutely correct. But Mr. Tock (for lack of a better name), didn't call me on it, now did he. :)
Your post opens up a whole new line of questioning, though.
As a supplement to the paper, this presentation offers some additional insight. According to the presentation, the use of a mid-gap metal reduced the required doping levels. Perhaps the reduction in doping levels was considered worth the performance hit from not using a matching metal? You would be trading performance for thermal budget. And thermal budget is going to be more and more valuable as the geometry continues to shrink.
As you pointed out, this is a gate first flow with a mid-gap matching metal. Is it possible that this is the 2nd generation HK/MG that intel has told us will be found on 32nm. I'm inclined to agree with your analysis that it is unlikely, but I don't think we can rule it out completely.
Another question this structure raises is what is a 32nm transistor? With a traditional planar transistor, this was (nominally) the gate width. But when you go to tri-gate, you have the gate in contact with the fin on 3 surfaces. Which feature is now the nominal 32nm feature. Does the fin structure allow you to go with a larger feature width, making the fill issue easier to resolve?
In short (or is it too late for that), there seems to be a wealth of information on the topic, and still no clear cut answers.
I'll answer the easier question:
Which feature is now the nominal 32nm feature.
Typically the node name is not the size of the transistor, but the metal 1 half pitch. This it essentially the minimum spacing between metal 1 lines (it is actually the distance from the middle of 1 line to the middle of an adjacentline, I believe). Traditionally for the last few generations the transistor gate length is smaller than the node name, I would suspect 32nm will be similar with or without tri-gate (the gate last flow now confounds how this is measured/quoted)
If I had to bet I would guess 32nm would be the next gen of highK/MG with traditional (planar) transistor and 22nm would be tri-gate.
This also brings in the question of SOI/tri-gate as SOI enables a lot more flexibility with well doping and channel and may give a lot more flexibility for a gate first/midgap metal (I believe SOI enables much lower well doping levels). Intel also is probably trying to tie SOI with some of the alternative SRAM technologies under investigation. I'm purely guessing/speculating on all of this.
That's a very good link though - one thing mentioned was that with lower channel doping needed for tri-gate, you get increased mobility which means a faster transistor, this is essentially a 'free' added benefit.
It would also eliminate HOT (hybrid orientation technology) substrates. I think this was first introduced by IBM - carriers (especially holes) are faster in certain crystal orientations and by making your pmos devices in this orientation you would get more transistor speed and you would leave your other nmos device in the standard orientation (where electrons are faster). As with everything, this means a few more process steps and added cost.
With tri-gate as you are now working on 3 planes (instead of just 1) you can now have your pmos device working in 2 "fast" planes and 1 "normal" plane which completely eliminates the need for HOT and the extra processing it requires. In a bit of irony, I think the double gate tech proposed by IBM does not do this and therefore the HOT would still potentially be needed/useful (not that this is a good thingm as again it means more processing and added cost and Intel essentially would be getting most of this benefit for free with trigate)
And the stress portion of the paper just makes my head hurt as now you have to think about stresses in 3 directions as opposed to 1, but it appears as though the trigate benefit is additive to the current Intel stress/strain technologies.
As for your mid-gap question - don't know! Maybe the benefit of tri-gate is substantial enough for Intel to "eat" any performance hit for using a midgap metal (also by the time trigate is implemented, who know there may be dual metal gate first solutions?). Also as you mention the difference in channel doping may help lessen the hit you take. Keep in mind a lot of Intel's early papers on a planar high K device were also using mid-gap metals as well - it is easier to process and allows you to do the research and get the major feasibility issues solved quicker.
Ahhhh, that's much better, tri-gate, FIN, channel doping, I think a cigarette after that.
ITK-The blue Logo was for mass consumption. After all, you know how modest I am about processors. :)
Orthogonal-
“What's in a name? that which we call a rose
By any other name would smell as sweet;”
(Romeo and Juliet)
Don’t worry about what they call ‘em. You just make ‘em 20% faster than my current bad boy, and they can call ‘em "iPalooka" for all I give a hoot.
iPalooka Extreme Edition!
Hoo Ya!
SPARKS
Gotta agree with ITK... Sparks - you call yourself an OC'in, watercoolin',self respectin' computer builder and use that logo?
Now some of us (like me) can't afford it, but it is truly a dark day... mass consumption or not!
"Intel Core i7 CPUs sport energy saving feature"
http://www.eetimes.com/news/semi/showArticle.jhtml;jsessionid=RXRWVJNB3YS4QQSNDLSCKHA?articleID=210000564
A company spokesman said it is not a direct evolution of the Intel's SpeedStep technology that automates frequency scaling based on workloads.
Any guesses before the IDF info release?
My guess is an on chip (or in package via MCM) voltage regulator. Intel had talked about this in the past and it would allow much faster voltage adjustments than the current solutions (which is voltage regulation on the motherboard). Though I doubt my guess is accurate. This could also have the potential negative impact of killing voltage OC'ing unless Intel unlocks it - which they probably only would on extreme chips, but would they on standard chips?
"If you find any issues with the content of this site or feel offended, please email your ISP and request for your internet disconnection"
"I'll answer the easier question"
No answer the hard ones. The easy ones are easy, tell us something we don't all know.
Like what does the future hold for AMD.
When will AMD really start production on 45nm.
When will we see AMD's first HighK Metal Gate
Or will they still be here for the London Games?
Tuesday, I purchased a copy of ‘CPU’ magazine. I don’t if anyone was aware that during AMD’s 2005 and 2006, shall we say heyday, they did reports on Athlon practically on a monthly basis. They would explain how it bested INTC’s offerings this way, that way, with power efficiencies, etc.
The magazine was fat with ad’s, from AMD, and 2nd tier vendors.
Thing’s are different these days, however. The AMD ad’s are gone. Intel bashing has given way to AMD’s graphics miracles and there are no INTC ad’s at all, let alone reviews. The reviews were dry, very dry, almost clinical. The mag is thin, very thin.
Which bring us to this month’s review of the top“Luxury Rigs” entitled, ‘Leaders Of The Pack’. They got the big named vendors’ rigs out for a $15,914 to a paltry $5,599 shootout, all 13 of them.
What 10 of them had in common was a QX9770, the rest:
1 C2D E8500 (4.6 GHz)
1 QX 9650
1 Skulltrail Monster (More of a media machine than Gaming Rig)
What struck me odd was there were no accolades, no reviews, and no commentary on the conspicuously absent Pheromone or, more appropriately, the exceedingly powerful QX9770.(My love, OXOX) There were no tongue-in-cheek comments, or any INTC bashing at all, nary a whimper. Even Mike Magee, former AMD pimp extraordinaire, was uncharacteristically muzzled.
I wonder why? They call it ‘CPU’ magazine, after all.
SPARKS
No answer the hard ones. The easy ones are easy, tell us something we don't all know.
Mr Tock, or should I say Lex, let's take a look at the old profile...
Male, 38, list Pornstar and Part time IT consultant as occupation (Lex is apparently sort for Lexington Steele according to the bio). Kind of ties in with the bodily fluid references, both on Scientia's blog under "Lex" and similarly on this blog with the Tock tagline, and the recent "crafty" comments without the tagline. Oh and happens to reside in AZ, doesn't Intel have a manufacturing facility down there? (Surely a mere coincidence) I do not mean to offend others but this apparently is a disgruntled / jealous AZ worker who doesn't like being a small fish in a big pond and is threatened by folks with some background that might reach a little deeper then his.
I'm not an AMD fan and as I have said in the past I frankly don't care if AMD goes under. The questions you posed have been discussed here already - it's just that at the time you were busy getting banned over at Scientia's, and weren't posting your witty comments over here.
So, thanks for dragging this blog down.
Robo - you've run a good blog and have been patient with folks, unfortunately thanks to clowns like Lex/MR.Tock/eventual new tagline/clock theme/AMD finished/BK/etc, I'm outta here. I'm sure Mr Steele will help keep the level of comments on this blog informative and of substance, when he's not insulting people and talking about bodily fluids.
Sparks... I've enjoyed your comments and enthusiasm! ITK - it was good to have the process discussions with someone without a chip on their shoulder and willing to have a give and take - I learned quite a few things from you. HYC - keep up the comments and keep contributing on the architecture-side as you seem to have a unique knowledge in that area that many folks here (especially myself) do not have.
And again Robo, thanks for facilitating the forum, and I hope folks like Mr Tock don't completely turn this into just another "lowest common denominator" site and force some of the 'good' (for lack of a better word) readers/writers out of here.
It's been fun gents!
Well, that's a shame to lose you Guru, but I can't say I blame you. I've enjoyed your stay as much as the rest. It's unfortunate that one individual can crash the party and take things to where they are now.
If Lex does work for Intel, which I hope he does not, but if he does, I'm gonna kick some ass.
Yeah I'm outta here to..
Who is this anonymous silly guy that throws stones but hides
Guru too funny
What has this monkey done to prove he is anything but a pussy?
How does Arizona kick ass?
Guru, it is a shame to see you go. I've enjoyed the conversation as much as you have, if not more.
I was particularly looking forward to getting your reaction to IBM vs. TSMC for Low-Power 32 nm.
Beyond the technical aspects of the article I found several things interesting.
First, they were gunning for TSMC. Does this mean that they don't consider Intel a contender in this space, or that they have thrown in the towel?
Second, this announcement is for 32nm. What the heck happened to "late 45nm" for HK/MG? That is the process the AMD faithful are holding out such hope for. AMD needs HK/MG, because to quote Scotty, they "cann'a change the laws of physics."
Third, they talk about customers "skipping" the 45nm node altogether. If that isn't an admission of the failure of a process, I don't know what is. You don't make that kind of R&D investment and "choose" not to sell it.
And this failed process is the one that AMD is stuck with. I have to wonder what Barcelona would look like on a good process. AMD has taken a lot of heat for Barcelona, and deservedly so, but this makes me question how much of the problem really hits back to design.
Finally, they make it sound like using strain techniques are a bad thing, citing added cost and complexity. The article doesn't address the issue, since it only deals with comments from IBM, but if the IBM process is using SOI, I think the whole cost and complexity argument is a wash.
“I hope folks like Mr Tock don't completely turn this into just another "lowest common denominator" site and force some of the 'good' (for lack of a better word) readers/writers out of here.
It's been fun gents!”
Whoa! Hey, no so fast!
I am VERY disappointed. By you leaving, you have invariably set the course for that which you say you hope doesn’t happen here by default. Ironically, you have enabled this effort by leaving.
“Guru too funny"
What has this monkey done to prove he is anything but a pussy?”
Listen, asshole, he’s made me fucking happy.
You want to know about pussy’s tough guy? Let me know when you’re in New York City. Just drop a line here; I’ll meet you on the corner of 51st street and Madison Avenue, SE corner, any weekday morning, between 7:45 and 8:00. I’ll dock myself an hour for the pleasure. The only thing that will beat you the hospital is the headlights of the ambulance you’re in. I gave GURU that name, DICKLEX.
"How does Arizona kick ass?"
Not a clue, douche bag, but you bet your pussy ass I know how New York City kicks ass. Let me give you an education.
You don’t make me happy.
GURU- Please, don’t leave.
SPARKS
AMD LOSES MORE MARKET SHARE TO INTEL:
http://www.eetimes.com/news/latest/showArticle.jhtml?articleID=210004369
HYC, I don't know if this is in your area of expertise or not, but it can never hurt to ask.
ATI/NVIDIA have both gone with fixed hardware implementations for handling graphics. Larrabee looks to be handling much more of this load in the software.
Being fairly clueless about architecture, I have to wonder a couple of things. Are graphics sufficiently parallel to take advantage of the multi-core approach that Larrabee will be using? Is the flexibility that software provides going to be able to offset the speed that fixed hardware will allow?
InTheKnow said...
HYC, I don't know if this is in your area of expertise or not, but it can never hurt to ask.
ATI/NVIDIA have both gone with fixed hardware implementations for handling graphics. Larrabee looks to be handling much more of this load in the software.
Being fairly clueless about architecture, I have to wonder a couple of things. Are graphics sufficiently parallel to take advantage of the multi-core approach that Larrabee will be using? Is the flexibility that software provides going to be able to offset the speed that fixed hardware will allow?
I don't have enough direct experience with the current GPU market to really say. My experience in the past is that dedicated hardware that you design today can be faster than anything you can do in software today, but the general purpose processors of tomorrow will make up the difference. And overall, coding for a general purpose processor is cheaper than redesigning your dedicated hardware year after year.
Some companies still build dedicated hardware for their projects though, because it gives them faster answers "today", even if the hardware will quickly become obsolete. There's always a window of time where the dedicated hardware is clearly ahead, before the software side catches up.
Guru: thanks for the interesting/challenging conversations. If you ever decide to pipe up on any other public forums I'd be interested to see. I learn a lot more from having my mistakes pointed out to me, than just having folks agree with me. I've learned here, from you.
"All you have to do is write one true sentence.
Write the truest sentence you know." - Ernest Hemingway
After more than a half a century on this planet, it never ceases to amaze me how one egocentric fool can cause so much calamity, the world is replete with them, and their actions are tantamount to social cancer.
What is the driving force behind this madness? I don’t know. Each participant has his/her issues, most are pathological, and all are socially destructive. Their methods however are quite clear. Disrupt a free flow of ideas with biased personal attacks, while satisfying some selfish inner desire, mostly low self esteem at the expense of others.
What I do know is to never enable the bastards. When we do so, we are all weakened as a group, collectively.
I also know when to recognize a loss.
If one considers Doc ROBO to be the soul of this blog, I think most of us could consider GURU to be its heart. He is kind, giving, and unselfish. A true intellect, his knowledge, academic, goes beyond conjecture or debate.
I seriously doubt in my life I will ever have the pleasure knowing, let alone developing a rapport, with an individual of such remarkable, expertise, character, and caliber. Someone, I would have liked to call my friend.
I hope it was half as difficult for him to leave as it was for us to see him go.
I shall miss him immensely.
SPARKS
Sorry to see you go Guru. Your posts here were always informative and insightful and I learned a lot from reading them. You'll certainly be missed by the majority of us here.
Sparks, presumably you've seen several reviews of the 4870X2 card. In particular the TechReport article ( http://techreport.com/articles.x/15293 )
What a total beast! 2GB GDDR5 on a single card, this thing was made for gaming on 30" displays. I picked up one of these: http://www.powercolor.com/eng/products_features.asp?ProductID=2351
There is some hilarity in knowing that such a card is being run in an nForce 790 board. Much faster than my dual 8800 GTs. Call of Duty 4 at 2560x1600 with 16X AF and 16X AA and all the details cranked up to full is quite a thrill!
Hey “G”, how’s it going, bud!? The release of the 4870 X2 has been everything I expected it to be, and more. As you know, I have been using a Crossfire setup for years. This new card is scaling up quite well. Crysis scores, however, have been a bit of a disappointment. I suspect it’s a driver issue. Time will tell.
Powercolor is going to release an overclocked version, 800MHz from the default 750. That’s the one I’m waiting for.
All said, the performance is excellent considering I paid twice as much for my 1900XTX setup a few years back. Combine this new card with the X48/DDR3-1800/QX9770 combo, and I’ve kept this upgrade well under 3 grand. Not to shabby, considering the entry fee to the “High End” vendors prices which start at 6 and end at close to 15! NFW!
And, I kept the whole shebang under my wife’s financial radar!
QX9770-1500
P5E3 Pre. 450
Patriot 1800- 300
4780 X2 -600
Monitor not included, that’s next. 1900X1200 is the target.
SPARKS
Well the numbers are in. Nehalem @ 2.93 has SMOKED my QX9770 (at stock speeds thank god). GURU was right again ~ “20 to 25%” as he put it. The first page of the article at Hexus is titled, “Bringing on the hurt”, and how!
What a monster!
http://www.hexus.net/content/item.php?item=15015&page=1
BTW: Super Pi didn’t mean squat to this bad boy.
SPARKS
In a bit of a surprise, IBM announced they have produced a functional 22nm SRAM cell. To the best of my knowledge, Intel has not done the same.
If, and this is a big if, IBM can successfully integrate this into a full blown processor within the next 3 years, this would put them on pace to catch Intel at 22nm.
Sparks said...
What is the driving force behind this madness? I don’t know. Each participant has his/her issues, most are pathological, and all are socially destructive. Their methods however are quite clear. Disrupt a free flow of ideas with biased personal attacks, while satisfying some selfish inner desire, mostly low self esteem at the expense of others.
I totally agree with you, but unless Robo is willing to spend more time moderating these comments, which is probably not practical for a part-time blog, then it is something we have to put up with. I prefer Robo's style of letting everyone freely post, and then only deleting the most egregious and worthless postings, rather than Sci's refusing to allow anyone to post anything without him first vetting and editing everything that is not 110% pro-AMD, which is what Sci does on his blog. No wonder it is nearly as dead as Sharikook's blog.
There's really no need for all the sex-based innuendo, and it looks like the anonymous poster (aka Lex as some have suggested) did clean up his act a bit, but most likely that would have been a short-lived improvement. What some posters fail to realize is that some of us read this blog from work (hopefully only on our lunch hour :), and the last thing we need is a bunch of X-rated material on our screens when the boss happens by, or CIO monitors what websites we're viewing on company time & equipment.
It's not possible to please all the people all the time of course, and one person's "too lenient" is another's "too heavy-handed" when it comes to moderation.
As an example of too much bad moderation, just look at Sci over on AMDZone: http://www.amdzone.com/phpbb3/viewtopic.php?f=52&t=135444&st=0&sk=t&sd=a&start=25
Here you have a mod (Sci) who complains constantly about all the ad hominem attacks on him here on Robo's blog, and then proceeds with the belittling insults on a "pro-Intel" poster who disagrees with Sci's viewpoints:
"Did you get this from your ouija board or tea leaves? Or is this based on your unwavering faith in Intel?"
and
Do you have any idea what a fool you are making of yourself? You come across as a die hard Intel fan and nothing more. You haven't seen me making pronouncements about Shanghai versus Nehalem or Bulldozer versus Sandy Bridges and I would hazard a guess that my estimates of the performance are a lot better than yours.
Now if the poster Das Tipitz had posted anything remotely as insulting as the above, he would have been perma-banned...
Not only is Sci pretty much clueless on technology, he also makes a terrible mod :). There used to be a mod named Che last year on AMDZone, who got into some disputes with Sci. Apparently Sci threatened to take his postings elsewhere unless the site owners got rid of Che. Instead of telling him "Don't let the door hit you in the arse", they caved in and now Sci is a mod.
Anyway we should all be thankful that Robo is more 'hands-off' here :).
“this would put them on pace to catch Intel at 22nm.”
ITK- I recall some of you guys saying that it was much easier to shrink SRAM than it was to shrink logic. This is what occurred to me when I read about IBM’s new “breakthrough”. Then again, what the hell do I know?
I wish ‘you know who’ was here to discuss it with you so that we all could learn from you both. I’m sure there are subtle process variables that limit what you can do with logic at these levels. That wasn’t mentioned in the article, nor did I expect it to be. Besides, if IBM can do it, I’m sure INTC can, too.
Further, the question begs, can IBM do en mass and in volume? That, I think, is quite another story. After all, look what they said they could do with SOI, years back. AMD unfortunately bought it, hook, line, and sinker.
SPARKS
“Anyway we should all be thankful that Robo is more 'hands-off' here :).”
Moose- I concur, 100%. A ‘liaise fair’ approach has been the DOC’s signature. The overall quality of the blog has been exemplary, despite our most resent irrevocable loss.
Besides, I’m certainly in no position to debate his wisdom.
SPARKS
Sparks, here is the link to the original article.
Not to diminish the accomplishment, but it is clearly over hyped. It is often over hyped when Intel makes the announcement too. A functional SRAM cell is the first step on a very long road. A functioning SRAM cell consists of 6, yes a whole 6, functional transistors.
So as you astutely pointed out there is a lot more to manufacturing a processor than just making an SRAM cell.
I thought the article that came out a bit later from Ars Technica did a better job of putting IBM's accomplishment in perspective.
ITK- That link was much better than the one I read, which spelled the end of INTC in 2011. The Ars Technica article was far more objective.
Speaking of hype, INTC, not to be outdone by IBM’s miracle technological breakthrough, was doing some, shall we say, showboating of its own. (It’s more like the two 800 pound gorillas were beating their chests.)
Anyway, Ed at overclockers had an interesting article on INTC’s 22nM roadmap today. He seems to think programmers are incapable of writing code for all that multithreading/cores. I’d like to know ‘hyc’s’ take on this since he writes the stuff for a living.
http://www.overclockers.com/
“Ivy Bridge and Haswell”
Within the article there is a link to a Canadian site which details 2011 and 2012 respective Tick Tock for those years. Also, Pat Gelsinger seems to be holding an extremely expensive 45nM Nehalem Frisbee, it could be a smaller process, but I can’t read French.
And no, I’m not that astute, I had a great teacher.
SPARKS
I thought that "a nonny moose" was the person that was accused of being "Lex" on another forum? All of these anon posts confuse me. =(
re:Ed's points on CPU evolution- I thought that AMD/Intel were going multicore because they were having problems continuing to increase raw clock speed. I figure it's one of those things where the technology has to take a new path and software will follow over the course of time. Though that doesn't really solve the problem of dealing with code/applications that don't benefit from multiple cores.
INTC is like a runaway Maglev. Don’t even try to get in the way. Subsequently, I think IBM’s 22nM announcement was timed to deflate the impact of the 2008 IDF.
"Intel’s 6-core processor code-named Dunnington will be called Xeon X7460 and is expected to become available in servers beginning next month. Intel claims that servers based on the chip already have broken performance records, including an 8-socket 48-core IBM System x3950 M2 server, which became the first platform to break the 1 million tpmC barrier on the TPC-C benchmark, Intel said."
Did I say 2008 was going to be the great server assault? I guess now know what CRAY was looking at months ago.
http://www.tomshardware.com/news/intel-idf-memory,6195.html
SPARKS
Okay, I just can't let this go anymore. I am sick of this weak analysis of the atom processor. I went to the link Sparks provided for the Overclocker's site and read Ed's article.
In the process I also read his most recent article and all I could do to keep myself beating my head on the table and screaming.
This idea that Intel is afraid of Atom's success is non-sense. The chip was designed to provide Intel with good margins. So what if it encroaches on Intel's low end laptop sales?
I found a price on the T2390 of $70. The die size is listed as 111mm^2. An Atom is listed as $25mm^2. That means that Intel manufactures 4.44 Atoms for the same cost as 1 T2390. Packaging costs will eat into that a bit so let's say that Intel makes 3 atoms for the cost of a T2390. So if Intel sells an Atom for ~$23 they break even on margins. And this will get better with 32nm.
Those numbers are crude, but should be in the ballpark.
Then there is the simple fact that there are things the Atom just doesn't do well. If all you do is email, browse the web and type the occasional document it is probably fine. Anything more than that and you need to buy a better processor.
So even if 60% of the users Ed is talking about don't buy anything more than an Atom, Intel is fine.
Tonus said...
I thought that "a nonny moose" was the person that was accused of being "Lex" on another forum? All of these anon posts confuse me. =(
Huh?? I might slam Dementia & Ah-Ben-Stoopid occasionally (OK, frequently :) but I don't use X-rated language doing so, and also I don't do the "tick tock of doom" for AMD either. AMDZonerz, yes but not the company. In fact, I think they'll gradually turn themselves around and reduce their losses (although not as fast as I thought). Anyway, I haven't seen any allegations that I am the dastardly "Lex" poster or imposter, until now that is.
I went with "a nonny moose" as a joke response to all the "anonymous" poster complaints here. Maybe I should change to Bullwinkle J. Moose instead :)...
http://www.eetimes.com/news/semi/showArticle.jhtml;jsessionid=PY44LD2PJY5HSQSNDLPSKHSCJUNN2JVN?articleID=210102280
According to Rattner, SOI will enable Intel to build surface devices like Fin-FETS and Tri-gates, that wrap 3-D gates around thin-film channels atop a SOI substrate. He said Intel would likley turn to SOI at the 22-nm node.
Damn, if Guru wasn't ahead of the curve on this one too. He said SOI and tri-gate would potentially be coupled and that while 32nm was a possibility, 22nm was more likely.
So glad we had that Tick Tock idiot drive him from the site and then fade away as he really has nothing to offer.
This was one of Guru's last technical comments, a little more than a week ago:
If I had to bet I would guess 32nm would be the next gen of highK/MG with traditional (planar) transistor and 22nm would be tri-gate.
This also brings in the question of SOI/tri-gate as SOI enables a lot more flexibility with well doping and channel and may give a lot more flexibility for a gate first/midgap metal...
“Damn, if Guru wasn't ahead of the curve on this one too. He said SOI and tri-gate would potentially be coupled and that while 32nm was a possibility, 22nm was more likely.”
Not only that, but he straightened me out when I brought out this link:
“SOI refers to the use of a layered silicon-insulator-silicon substrate on which transistors are built, rather than a simple (bulk) silicon substrate. Some companies claim to get some performance and/or power benefits from SOI over bulk silicon. Intel’s analysis shows that such benefits, if any, are marginal, and do not justify the substantial cost increase of SOI wafers. Intel has never used, nor does it plan to use, partially depleted SOI (PD-SOI) that others are using. There is another type, however, called fully-depleted SOI (FD-SOI) that is under investigation at Intel and is not being used by any chip makers today.”
http://www.intel.com/pressroom/kits/45nm/IntelHigh-K_metal_gate_glossary_FINAL.pdf
The glossary term is under the heading “related terms” on page 3 of the .PDF.
He called my post bit of speculation coupled with a bit of knowledge, or something to that effect. To me, this was a compliment of the highest magnitude. To him, in other words, I got lucky. (Hey, coming from him, I’ll take what I can get!)
Obviously, INTC had left the door open on FD-SOI and GURU had peek inside. At least, he knew the FD-SOI was viable alternative technically, and pegged it. He also knew PD-SOI was a dead end, citing leakage, Ion and Ioff being the critical criteria, a year ago. Brilliant.
I’m still hoping he returns.
It feels like a death in the family.
SPARKS
Moose- I think TONUS wasn’t making an accusation. I believe he got confused with all the anonymous posts. I got caught up with this once or twice myself, and said some stupid things to GURU, no less! He knew I got confused and gave me a pass. After that I took a profile.
No one ever would confuse you with DICKLEX. Trust me.
SPARKS
ITK-
“This idea that Intel is afraid of Atom's success is non-sense.”
I agree.
Ed, for the most part, has been pretty accurate. This time, however, he got a bit carried away. INTC can’t crank out enough of the little beasties. I suspect it has to do with the Israeli FAB delay. I’m guessing, of course.
http://www.theinquirer.net/gb/inquirer/news/2008/08/21/atom-shortage-hits-asus
SPARKS
Hey “G”, how’s it going, bud!? The release of the 4870 X2 has been everything I expected it to be, and more. As you know, I have been using a Crossfire setup for years. This new card is scaling up quite well. Crysis scores, however, have been a bit of a disappointment. I suspect it’s a driver issue. Time will tell.
I've been quite busy with work as of late. :-( Pay might be alright, but some of the deadlines are sheer insanity! How are you doing? Finding plenty of free time to enjoy your monster system I hope. :-)
You're right about the Crysis scores. I tried for myself and at 2560x1600 I was getting lower frame rates than with a pair of 8800 GTs. I suspect as they work on the drivers the scores in Crysis will dramatically improve. The scores in other games such as Call of Duty 4, Oblivion and so on are just unmatched by anything Nvidia has to offer.
Powercolor is going to release an overclocked version, 800MHz from the default 750. That’s the one I’m waiting for.
Nice! A word of warning though, this card runs hot. Extremely hot. I'm actually contemplating picking up a water block for it.
And, I kept the whole shebang under my wife’s financial radar!
QX9770-1500
P5E3 Pre. 450
Patriot 1800- 300
4780 X2 -600
Monitor not included, that’s next. 1900X1200 is the target.
If you can sneak it by your wife and the budget allows, I'd highly reccomend a 30" display. I've got the 3007WFP and it's just goregous. Gaming at 2560x1600 is just absolutely incredible. The panel is S-IPS, so it's one the best LCDs on the market.
BTW, my list is even worse than yours!
E8400 - $200
790i - $350
QX9650 - $700
Dual 8800 GT - $400
3007WFP - $1400
DDR3 - $300
Koolance Water Cooling Gear - $350
I've been playing with OCing the CPU some more as well, and the results speak for themselves. 457 x 10 for 4.57GHz gave a SuperPi time of 9.8 seconds! Very happy about beating 10 seconds finally. Of course, the AMDZONE members will claim that SuperPi doesn't matter. In fact, I think SuperPi only stopped mattering to them around the launch of Core 2 Duo. An amazing coincidence isn't it? Obviously I wouldn't run the system at 4.6GHz full time! 4.275 is more than enough. :-)
Well the numbers are in. Nehalem @ 2.93 has SMOKED my QX9770 (at stock speeds thank god). GURU was right again ~ “20 to 25%” as he put it. The first page of the article at Hexus is titled, “Bringing on the hurt”, and how!
What a monster!
Looks like I'll need to buy an X58 motherboard, another two sticks of DDR3 memory (6 1GB sticks, to make the most of the tri-channel memory controller) and a Bloomfield CPU as an Xmas present for myself! At AMDZONE they're in a state of denial. The benchmarks are all fake and so forth. I remember seeing this before. When was it? Oh yes. Right before the Core 2 Duo launch! Seems that history is repeating itself!
-GIANT
Forgot the 4870X2 off of my little list. That's a nice $580!
SPARKS said...
Moose- I think TONUS wasn’t making an accusation. I believe he got confused with all the anonymous posts.
Yeah, I think Hyc did too :). Usually you can tell the anonymouses (anonymii??) apart by the writing style after a while, however. I too hope Guru comes back as well as JumpingJack (haven't seen him in a while).
Guru said...
Monitor not included, that’s next. 1900X1200 is the target.
I got a Dell 2707 aabout 18 months ago and have been really happy with it. Widescreen, wide color gamut (92% I think), and fairly fast response time of 6ms GTG. First one went bad within a week but Dell shipped a replacement the same day I called. 1920 x 1200 and great for gaming. Now Dell offers the 2709 which has even wider color gamut and cheaper too :).
So I keep seeing the question what is "Joe Sixpack" is going to do with all these cores?
I would offer speach recognition as an option.
If we wade through all the marketing fluff, there are a couple of interesting things here.
First we have
The results of the implemented optimizations are clearly visible: Dikte’s
performance scaled by a factor of 3.2 from one to four cores with
a recognition accuracy of over 95 percent.
So the software seems to scale at about 80% per core. Realistically, scaling probably drops with the addition of each additional core added, but up to a point there is still a benefit from more cores.
So the addition of more cores would allow word recognition to become more accurate and to work more quickly. To be more useful, the software would need to recognize the speech patterns of multiple users. That of course means even more computational power.
And then the article goes on to say
The maturity of the recognition engine also means that it can easily be
adapted to any other human language; adaptation of Dikte to recognize English has already started.
It doesn't take much of a stretch to see this moving to real time translation.
Now imagine an 8 core atom manufactured at 22nm in a MID sized device that can run this software for 8-10 hours on a charge and you've really got something.
That is what I believe the potential is for atom and why I think it is such a big deal.
Giant- Bro! Crysis at 2560X1600 on a 30!?!? I’m surprised the thing is even playable! Man, you don’t ask for much! Crysis is THE benchmark by which all, and I mean all, graphics cards are judged, be it 280’s in SLI or 4870 X2’s in Crossfire X (2 cards).
You can’t even run “Very High” settings in Crysis without VISTA. Misrosoft (sic) decided, as a marketing scheme to get gamers to jump on board, they would make DX 10.1 incompatible with XP PRO, bastards. No Vista, no DX 10.1. Well, as far as I concerned, they can shove it. I certainly didn’t buy all this expensive hardware to satisfy MS’s view of the bloatworld, simply to obtain the highest resolutions in a hand full of games.
In any case, I can’t imagine ‘Crysis’ at “high” resolutions on a 30. This truly must be something to behold.
There’s NO WAY I could slip a 30 inch bad boy past the little lady. We wouldn’t have “dinner and a show” for a month, if you know what I mean.
The Samsung 245T has a 3.5 (out of 10) on the WAF (Wife Acceptance Factor). It would be a mild piss off, but a nice bribe of some sort would smooth things out. A “Day Of Beauty” would be the weapon of choice here.
The 245T has an NSTC color gamut of 97%, is 8 –bit, and is LED backlit for superior contrast and grey scale. Further, at around 700 bucks it’s well below the WAF radar threshold. Further, a 4870 X2 @ 1900 X 1200 would have excellent frame rates in Crysis @ “High” settings.
DELL 3007!
Hoo Ya!!
SPARKS
BTW: I miss GURU.
Moose-
“Now Dell offers the 2709 which has even wider color gamut and cheaper too :).”
Thanks for the tip. I’m going to compare it with Samsungs 245T.
ALL- Here’s some nice juicy links for your Holiday shopping pleasure. At the risk of sounding redundant, what a monster!
http://anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3382
http://www.tcmagazine.com/comments.php?shownews=21369&catid=2
This thing is going to sweep everything else aside, bar none!
SPARKS
Here we go again.
Last week AMD stock rose substantially, but inexplicably. I didn’t get it until I read the first link ROBO posted on this blogs opening page: ‘Analyst Gays AND Gaining Ground’.
However, this Anal-ist didn’t disclose everything he thought, neither did the article. Subsequently, I read more articles and then found this piece:
http://www.clusterstock.com/2008/8/amd-selling-some-chips-finally-but-running-on-fumes
What wasn’t said in first article is that this SAME anal-ist also thinks AMD will run out of cash in 6 months! WHAT!?! He is calling AMD a “market perform”. Huh? Does this mean that AMD will outperform by losing less money than expected? Yeah, sure, this looks good, like a screen door in a submarine.
AMD has NOTHING competitive to offer.
Nehalem is looming over AMD’s head like the Sword of Damocles
No FABS, no R+D, no money, no 45nM, and no road map.
Any profits from ATI are being siphoned off to pay AMD’s debt.
AMD current Cripple Triple’s and Pheromone are power hogs.
Q6600 is STILL a better buy to this day, performance wise.
AMD is still 5B in the hole.
INTC will be making scheduled seasonal price cuts.
But this knucklehead feels, "increasing yields and shipments on triple- and quad-core products," which has led to "a small bit of momentum shifts back from Intel".
His name is Craig Berger of Friedman, Billings, Ramsey & Co. Can anyone tell me what color is the sun on this guy’s planet? I think it may be green, green enough to kick AMD’s stock up a buck or so in a week. AMD is currently at 5.80.
His target price is five bucks.
Huh?
SPARKS
sparks said:
Anyway, Ed at overclockers had an interesting article on INTC’s 22nM roadmap today. He seems to think programmers are incapable of writing code for all that multithreading/cores. I’d like to know ‘hyc’s’ take on this since he writes the stuff for a living.
There's a lot of truth in that. I've been writing parallel/multiprocessor code since ... 1987 or so. A lot of computers have come and gone since then; a lot disappeared because they were "too hard to program"... Personally I like the challenge; I spent an entire weekend shivering in a lab at JPL disassembling the TCP driver of an Alliant FX/8 to figure out what it was(n't) doing, and managed to double its performance on our pair of 8-processor machines. (It wasn't my assignment, I was just doing it for fun.)
The thing is, sooner or later you run into Amdahl's Law, which roughly states that any computing task may be broken into portions that are parallelizable, and portions that are inherently serial. And the relevant point here is that beyond a certain level of parallelism, the performance is totally dominated by the serial portion, after which point additional parallelism yields no performance gain.
Networked servers (like Apache, and the LDAP server I work on) ran into this problem in a very visible way - they typically had a single thread listening for network traffic, and then dispatched incoming requests to a pool of worker threads. This worked OK as long as your number of simultaneous requests was in the hundreds. But if you ran that server on a monster machine with thousands of processors, and expected it to handle tens of thousands of simultaneous requests, you found that your single listener thread becomes a serious bottleneck. And so you had to redesign your listener/event drivers to allow greater concurrency in logic flows that used to be purely serial.
Doing that redesign can be a real pain; you start needing to add locks to all kinds of data structures that didn't need locking before, etc... It's easy to do this wrong, it's hard to debug when you get it wrong, and it's hard to actually prove that you did it right. It takes a great deal of obstinacy/persistence to get it right. The average programmer doesn't have what it takes - they work for their paycheck and clock out at the end of the day. You have to fanatically love this kind of work, regardless of the hours or the pay, to get it right.
But also, the average software problem hasn't really needed this approach before... So yeah, there's a certain amount of training lacking out there, but also IMO there's a lack of aptitude. And finally, there's a hard wall (Amdahl's Law) where diminishing returns just makes it pointless to try that hard in a lot of cases.
Looks like asset light is really taking shape now. AMD has divested itself of plenty of assets in the last couple years.. First Spansion, now the Digital TV Market. They've also sold off old 200mm equipment, a large stake to the Arab's and I think even a little bit of real-estate in Sunnyvale.
There's not a whole lot left to sell to generate a positive cash flow. I think the analyst Spark's quoted may atleast be partially right that AMD may reach $800M in available cash by middle of next year.
Sparks said...
What wasn’t said in first article is that this SAME anal-ist also thinks AMD will run out of cash in 6 months! WHAT!?! He is calling AMD a “market perform”. Huh? Does this mean that AMD will outperform by losing less money than expected? Yeah, sure, this looks good, like a screen door in a submarine.
Ed at Overclockers has an article on Deneb, quoting Fudzilla (not a safe practice usually :) as stating many large clients have yet to see any engineering samples. According to Ed, if AMD can't get samples out to their large customers (mobo makers for one) in the next few weeks, we won't see any Deneb systems in Q4 despite AMD stating they "are on track for 45nm".
In fact, Ed says as much as 94% of AMDs product line is still K8. I personally find this incredible seeing as how decent (i.e., bug-fixed and clocking above 2.2 GHz) Barcies and Phenoms have been available since 5 months ago. I'd really like to see where Ed gets his info - hopefully it's not user comments on Newegg.com (where Sharikook gets 90% of his factoids).
If this is true, then 65nm was truly disastrous for AMD (terrible yields I would guess). And unless we see some hard evidence (demoes or widely-available benchies), it could be that 45nm is not doing as well as AMD would lead its investors to believe...
HYC-
“The average programmer doesn't have what it takes - they work for their paycheck and clock out at the end of the day. You have to fanatically love this kind of work, regardless of the hours or the pay, to get it right.”
Much can be said about these “paycheck” people, they’re in every facet of “professional” life. Frankly, I don’t enjoy working with people with this kind of mindset.
During the early years of my morning commute, I had the pleasure of knowing a gentleman who was a programmer for a large bank. I may have mentioned him before. He left me in awe. Gary was a large, quiet, soft spoken, kind man who wrote and dreamed code! Throughout his professional career he worked on IBM 360/370/3090/390 mainframes. (Like you, he absolutely swore by 64 bit and its efficiencies and abhorred the 32 bit environment, not to mention the translators and its inefficiencies, as they migrated to today’s MS NT servers.)
His bank recruited/hired a recent Harvard grad. This “kid” was a bit too cocky and arrogant for Gary’s taste as he was very modest and unassuming. He gave the new fellow an assignment dealing with some esoteric mortgage calculations. A few days later the “kid” came back with his submittal, Gary looked it over, and told him it was too long and sloppy, and tossed it back at him. Well, can you imagine the indignation? The Harvard kids’ masterpiece was rendered trash in a nanosecond! He was floored.
Luckily for junior, Gary showed him how to tighten up his lines, drop redundancy, and everything else you boys do to make an efficient, bug free operation. (Stuff laymen like me will never understand).
Subsequently, a few days later, the kid came back with something that was half the number of lines which Gary called, “beautiful and elegant”.
Gary, quite unfortunately, is no longer with us. However, before his passing, he told me they became very close and “The kid” went on to be a top executive who always managed to find time for an impromptu visit. After many years, he admitted that Gary inspired him to love his work.
That’s part of the training you mentioned, and it goes far beyond just being a job. Let’s hope for the industries sake, in this country, we have more people like you and Gary who will do their utmost in writing efficient code (out of pleasure?) to eliminate single threaded/serial bottle necks. After all, the hardware is ready.
I refuse to sanction Ed’s at Overclockers single handed write off of an entire professional group. I think with the right people, and incentives, we can do anything.
SPARKS
“If this is true, then 65nm was truly disastrous for AMD (terrible yields I would guess).”
It’s a good guess. What I would love to know, and probably going to go to my grave never knowing, is what the percentage of Cripple Triples are to fully functional quads are per wafer. After all, they did create a market for these losers out of dire necessity simply to obtain revenue from something that would be otherwise trash.
Perhaps, it was 20 to 30 percent? Who knows? The proof was the idea of the Cripple Triple can very late in the Barcelona game. The percentage had to be high no mater how many steppings they did, including the last one where they obviously threw in the towel and said enough.
That said, if it’s true about the 94% figures, I think no one, even those on this site, expected it to be THAT BAD! Therefore, it’s no wonder why INTC kept Q6600 on way past it product cycle, especially the cheap rocket (GO) stepping.
Frankly, I’m very surprised about that number. 45nM Denebs better be more than a “dumb shrink” (GURU hated that expression) as thinner layers will only exacerbate an already terrible situation. Could this be the reason for the Denebs delay?
SPARKS
Frankly, I’m very surprised about that number. 45nM Denebs better be more than a “dumb shrink” (GURU hated that expression) as thinner layers will only exacerbate an already terrible situation. Could this be the reason for the Denebs delay?
Actually, the 45nm *will* be a dumb shrink, although such a thing does not exist these days as in the past.
With immersion, AMD *may* be able to get by with similar design rules and layouts as their 65nm -- but from a defect perspective, this probably isn't a good idea.
If you look at IBMConsortium transistor data -- they *have* made improvements to the switching speed and (IIRC) brought down the overall leakage a bit. I wouldn't expect a repeat of the *worse* transistor performance 65nm compared to 90. There will be improvements.
Regardless, without HiK, AMD is going to get slaughtered in thermals. Or speed.
Actually, the 45nm *will* be a dumb shrink, although such a thing does not exist these days as in the past.
Let me clarify -- by a dumb shrink I mean that these days there needs to be a different circuit layouts/floorplanning -- the day of a pure optical shrink are long over. So, same design, with no/minimal new overall features.
“Regardless, without HiK, AMD is going to get slaughtered in thermals. Or speed.”
It’s amazing how INTC’s nuclear control rod alloy Hafnium based metals are fully implemented in a full production cycle while everyone else is trying to come to terms with INTC’s proprietary and obviously very secret process.
What I can’t come to grips with is how INTC takes 40 cent hit today while AMD sells its TV thing for a miserable 185M and stock increases 12 cents!
BTW: NVDA got slammed, too, despite a dollar a share increase last week after a Billion dollar stock by back. Not good.
AMD shares have gone up a buck and a half in one month. There’s more here than meets the eye. 185M simply doesn’t cut it. AMD slices off pieces of former ATI while siphoning off ATI profits to stay afloat. I suppose the 2006 purchase does have its rewards, after all, they can also write off billions each quarter while milking it for everything it's worth.
Something else is going on.
Then there is this from a well known AMD cheerleader:
http://www.fudzilla.com/index.php?option=com_content&task=view&id=9070&Itemid=1
I don’t get it.
SPARKS
Actually, the 45nm *will* be a dumb shrink, although such a thing does not exist these days as in the past.
To the best of my knowledge this is not true. If nothing else, Shanghai/Deneb will add additional cache. I would be very surprised if there were not some other tweaks made as well.
Penryn is the 45nm version of Core2, but they still added some minor features, etc. So while it is the same basic architecture, it is not the same processor.
Looks like SLI is coming to the X58 without the stupid bridge chip, as Anand reports.
Take it away Sparks!
Ah, perceptive and clever Chuckula, don’t mind if I do!
Where do I begin? Let me count the ways!
Last week NVDA bought back 1B in stock. They were down to 10.58 a few weeks back. The word got out and they got a significant up tick to 14.08. They are once again bleeding, currently at 13.14 while the rest of the market took off like a shot, just today.
They are now, like another company we all know, are mewling and puking the INTC bully rhetoric. In fact, what they did was commit themselves to a policy of a short term gain (LGA 775), as opposed to a long term strategy (Nehalem). Hence, they held on to their chipset/motherboard operations with tooth and nail fervor. Last year they made a fortune and scored record earnings with this, the “SLI only for us” policy, and graphics card dominance scenario. They had the entire industry by the balls. Now, with the introduction of Nehalem and no license to sell X58 chipsets, they are clearly S.O.L.
What exacerbating their problems is their notebook chips are literally melting off their solder lands! (They call it cracking) Horseshit, they screwed the pooch when it came to thermal design parameters. Incidentally, they initially tried to blame their partners , and then they tried to blame the process houses, in short, the collective backlash has knocked them off their collective feet. The bottom line you ask? They’ve set aside nearly a quarter of a billion to remedy these and future issues with all their partners.
Enter ATI, with low price, high performance, and relatively low cost X4xxx product lineup. Exclusive partners, like BFG and EVGA are going to ATI as NVDA dictated pricing policy with Gestapo like efficiency. I am quite certain AMD/ATI was more than eager to cut sweetheart deals with anyone who was interested. Everyone was. They all immediately jumped ship. NVDA’s extremely large (and no doubt expensive) GPU’s, margins were slashed dramatically. The previously mentioned partners (and others) weren’t about to eat the difference, NVDA took a bath. Further, the baby went out with the bath water, ATI/AMD has gained market share quickly and substantially, purely at NVDA’s expense. Hell, even via saw small, but noticeable, gains.
Motherboard manufactures are NOT interested in adding the cost and complexity of a dedicated SLI chip for future products for a very small market that actually use multi card setups. Basically, they said shove it.
Jen Sing Song is out beating the rhetoric drum with a “kinder and gentler” Nvidia in the Web tabloids. Now, he’s very accessible and very accommodating.
Bullshit, I say. Enthusiasts, AMD/ATI, the industry, and ultimately, INTC wants his balls in their top desk draws.
Call this the “Perfect Storm” 2, obviously, as you so astutely pointed out-----
“Take it away Sparks!”
I am enjoying every moment of it. I’m wiping the drool from my keyboard.
SPARKS
Chuckula
It gets better, ahh, worse?
http://www.theinquirer.net/gb/inquirer/news/2008/08/28/nvidia-55nm-parts-bad
Enter ATI, with low price, high performance, and relatively low cost X4xxx product lineup. Exclusive partners, like BFG and EVGA are going to ATI as NVDA dictated pricing policy with Gestapo like efficiency.
NVIDIA is not the first Taiwanese company to abuse their customers when the market is good and pay for it later. The PCB shop I worked for was Taiwanese and they did the same thing when the dot com bubble was growing. After the bubble burst, their customers took the opportunity to pay them back in spades.
That isn't to say that American companies weren't making hay while the sun was shining. But judging by the amount of ill will present when the shoe was on the other foot, I'd have to say that my company was worse than many others.
I assure you that management was not about customer satisfaction. That played a big part in my departure.
ITK- (I wrote this the other day but I didn’t post it, as not to hog the blog. But since it’s been you and me lately, after the passing of GURU. What do you think?)
I read Ed’s at ‘Overclockers’ piece. It’s all absurd speculation and nonsense. I quote---
“For the next year, for practical purposes, AMD will not compete against Bloomfields, but rather Penryns. That's still hardly favorable to AMD, but at least they'll have a fighting chance with 45nm chips against those, and still get a half-decent price for them.”
This is more ridiculous than his comment about INTC fearing ATOMS success. Currently, AMD is finding it difficult to compete with aging Conroe! Q6600 is and has been the biggest pain in the ass for AMD----EVER. To think, by any stretch of the imagination, AMD can compete with this monster that’s been cranking away @ 4 GIG, sitting in my machine since April, then a serious reality check is in order!
First they’ve got to get past this:
http://www.newegg.com/Product/Product.aspx?Item=N82E16819115041
They all seem to think that AMD’s 45nM will be as successfully implemented as was INTC’S! INTC made it look easy, too easy. Thanks to you boys, including my beloved GURU, there’s no way this “shrink” is going to be anywhere close to what they expect. In fact, I’m willing to bet they are having more problems at 45nM then they did with Barcelona at 65. That’s why they’ve been so quiet.
Hell, this is Deja Vous all over again. The same way Barcelona was going save the company last year, apparently all this hype surrounding Shangri-La and Deadbeat appears to running the same course.
WHERE’S THE BEEF! INTC had samples of Penryn last nearly a year ago. I see absolutely NOTHING of AMD at 45nM.
They need LN2 to get Pheromones to run at 4 GIG! LN2! Even with this the son of bitch is still slower than a stock clocked Penryn, clock for clock. Does anyone really believe that IF AMD releases ANYTHING at 45nM it going to compete with Penryn???
http://www.fudzilla.com/index.php?option=com_content&task=view&id=9082&Itemid=1
Ed may be smoking celery sticks, but he surely may right about these, however:
“Unless Deneb samples show up in Taiwan in the next couple weeks, they aren't going to show up (except maybe in one or two rushed quasi-baked mobos) in 2008, and that's bad, whatever the reason.”
And---
“It doesn't help you much to get a decent price on 4% or 6% Phenoms or Barcelonas sales when the other 94% of your sales are bargain-basement prices on X2s or lower.”
Sure, 6%, right, Crippled Triples, OK, 45nM, yeah right.
SPARKS
What exacerbating their problems is their notebook chips are literally melting off their solder lands! (They call it cracking)
Sparks, cracking is the right term.
Even if your component runs hot enough to melt the solder, the lands and the mounting holes are already wet with solder. It isn't likely to cause your connection to fail open from a single cycle.
What would happen is that over several cycles the component floats up out of the holes (it is less dense than the solder). after the component has ridden up out of the holes the solder solidifies before component has finished cooling. Without the support of the board around the lands, the stress from the component shrinking as it cools causes the solder joint to crack.
I know this is a technicality, but you know how the whole professional pride thing goes. :)
Sparks, I actually agree with him to a point. The chip AMD has to contend with in the PC space is Penryn from a price perspective. There is unlikely to be any competition between AMD's offering and Bloomfield.
I have serious reservations about AMD's 45nm process. All the evidence seems to indicate that AMD's 65nm leaks like a sieve with all the resulting thermal problems that brings. I haven't seen anything that indicates they have done anything to address the leakage problems on 45nm.
I'm sure that is because they don't want to tip their hand and have Intel steal something that will take Intel two years to work into their next design.
And then there is the issue of credibility. Based on the Barcelona launch, "on track" seems to mean that they think they will hit their current target until they "update" the target next week.
AMD could pull a rabbit out of the hat, but you'll have to forgive me for not holding my breath.
“I know this is a technicality, but you know how the whole professional pride thing goes. :)”
No sweat (forgive the soldering pun), but I’ve held a soldering iron in my hand a couple of times (I once built a Heathkit 25 inch TV set). I had no idea that this “cracking” was an issue in SMT, nor did I know the terminology. I always assumed SMT devices dissipated heat much better than more conventional components due their close contact to the board itself.
Now that you’ve explained the dynamics, now I’m sure it was a thermal design parameter overlooked, missed, and/or under engineered, by NVDA itself.
Thanks!
SPARKS
“AMD has to contend with in the PC space is Penryn from a price perspective.”
Ah, now that’s a horse of a different color. The key phrase is “price perspective”. You qualified what he didn’t say. As you know, price v performance is not something I think about too often. I assumed he meant on a pure performance level. To which I still say nonsense. It ain’t catching QX9770.
If they do, that will be a BIG rabbit from a little hat. No way.
SPARKS
Ed is usually looking at the business side of things, which is the side that I think many hobbyists overlook or just ignore altogether. While there is merit to studying and understanding the technical aspects of CPU design and performance, that alone doesn't determine market success.
AMD (and its followers) experienced this during the time that they had the performance lead. And they're seeing what happens when practical reality combines with Intel having the performance lead. All of the technical talk about compilers and performance on benchmark A or stability when overclocking with Prime95 has no impact on that.
Hey, let's discuss the technical merits of NVIDIA's GPUs, and ignore the recent problems with notebook graphics that are causing them some financial pain and lots of hand-wringing!
Sparks, I didn't see NVIDIA's cracking comment myself. I have to wonder if they were referring to barrel cracking rather than solder cracking.
Barrel cracking is an issue with the board not the components mounted on them. It is caused by issues in the plating process.
Either, the balance of chemicals in the plating bath is incorrect and results in the copper being brittle (rather than ductile). Or the plating didn't meet specs and was too thin.
Barrel cracking occurs when one or more of the interconnects (those copper holes with nothing mounted in them) that connect the traces on the various layers together cracks during assembly. This results from an open circuit.
If NVIDIA is claiming the issue is barrel cracking, then they are blaming the problem on their board manufacturer.
This is possible, but QC in the board shop should catch this before too many boards are affected. Both chemistry and thickness are monitored on a periodic basis and should be detected fairly quickly.
ITK-
Here are a couple of links for your review concerning NVDA’s mechanical falures. Perhaps you can shed some light on the topic/subject.
http://www.xbitlabs.com/news/video/display/20080730062939_ATI_Mobility_Radeon_Graphics_Processors_Packaging_Is_Solid_Says_Company.html
http://www.maximumpc.com/article/news/ati_says_you_wont_have_relability_issues_with_mobility_radeon
http://www.techpowerup.com/69704/NVIDIA_GPU_Failures_Caused_by_Material_Problem_Sources_Claim.html
Tonus- are you suggesting I stop beating up AMD? For you anything, however, I can’t give the tech press a pass when they print speculative assumptions as gospel. They’ve got to be called out on BS and paid marketing hype. After all, that’s why this blog was formed. 'AiMeD Corp.'
Beside, I love nothing better than to beat up NVDA. See above comments.
SPARKS
Tonus- Sorry, I couldn’t resist this one. Wrector Ruinz has got to be the Bill Clinton of the CPU world, powerless but a powerful presence.
He went to visit Luther Forest. Then he went to Saratoga to bet on the horses!
Does this give a new meaning to betting on the company? Hey, with his golden parachute he can afford to!
http://albany.bizjournals.com/albany/stories/2008/08/25/daily16.html?ana=yfcpc
SPARKS
sparks: "Tonus- are you suggesting I stop beating up AMD?"
Not at all! I was referring to people who base their opinions on technical aspects while ignoring the practical ones. It's all part of the same puzzle, and you can't leave pieces out. I know I took that same shortsighted view of the CPU/GPU industries years ago as well.
The discussion here tends to cover a lot more than just technology, but I still see where other sites seem to focus almost exclusively on performance. When they try to address other issues, it is often pie in the sky scenarios where government agencies wipe out Intel and hand AMD the world on a platter. To be honest, even the technical discussion isn't entirely realistic either, as it focuses on benchmark software, or uses examples that don't reflect real world use (ie, wondering how well a CPU would do if developers used different compilers).
For better or worse, the CPU design and development race does not occur in a vacuum. Ignoring the real-world factors might help keep fanatics warm at night, but at some point reality will come knocking. At the moment it is banging on AMD and NVIDIA's doors pretty hard. I may want those companies to do well, but I can't ignore reality when its tugging at my shirt sleeve.
Tonus said...
other sites seem to focus almost exclusively on performance. When they try to address other issues, it is often pie in the sky scenarios where government agencies wipe out Intel and hand AMD the world on a platter. To be honest, even the technical discussion isn't entirely realistic either, as it focuses on benchmark software, or uses examples that don't reflect real world use (ie, wondering how well a CPU would do if developers used different compilers).
Geez - if I didn't know better I'd think you were talking about AMDZone :).
At least Ah-Ben-Stoopid over there makes no pretense of being anything other than an AMD promoter (aka "Shill: "we are on a mission to promote AMD processors". Too bad Sci doesn't just give up his pretense of any impartiality and admit to the same...
It's kinda funny that Abi likes to go after the more senior posters on AMDZone, the ones who apparently have major business interests in rendering or company servers, and who voice their concerns over staying with AMD. I guess he's the self-appointed Sheriff gunning after everybody who fails to drink the Koolaid with every post...
ITK-
What do you make of this?
“Surprisingly, but the 45nm process technology will not allow AMD to reduce power consumption of its quad-core AMD Phenom X4 process considerably and the new chips are projected to have 125W”
http://www.xbitlabs.com/news/cpu/display/20080828043213_AMD_s_First_45nm_Desktop_Microprocessors_Set_to_Arrive_on_the_8th_of_January_2009.html
I wonder how the power reduction scheme used in 65nM Barcelona was a contributing factor in its low clocks at 65nM. Further, how would the differences in feature size @ 45nM prohibit power REDUCTION!?!?!.
Last year we spoke of these timing vs. power issues at length. This may point to the architectural failures inherent in the design itself, as we speculated. Perhaps the chip runs better when all cores run I/O synchronously. Wouldn’t this point to a problem with the memory controller addressing each core separately at different speeds or at different thermals?
The leakage issues, apparently, have only seen modest improvements from 65nM transition, roughly 15%. From the article they do mention a possible 3 GHz part. AMD at 45nM could, indeed, compete with lower end Penryns as Ed at Overclockers speculated, performance per dollar being the standard. But, they are going to get killed on margins.
It seems what they gained in the process shrink, they lost with higher clocks. This is essentially a wash, as they are back to 125W for a miserable couple of hundred MHz @ 45nM.
GURU’s speed limits on Barcelona seem to hang on like Einstein’s limits on the speed of light, even at 45nM. It seems no mater how they tweak, shrink, or step this thing they can’t get it to run fast, or efficiently. We’re talking physics here. I guess he was counting the number atoms regarding the thickness of each layer.
January 8th seem a bit far away when INTC is releasing cheep stuff like this now:
http://www.theinquirer.net/gb/inquirer/news/2008/09/01/intel-updates-processor-prices
All said, INTC just keeps the pressure on, relentlessly. They are not going to give AMD any respite at all, despite what Ed say’s on how Shanghi and Deneb will compete with Penryn. Frankly, where, and at what price point?
SPARKS
ITK- Charlie has an in-depth report of Nvidia’s mechanical chip failures. He’s calling them “bump cracks” which you mentioned earlier. Here’s the link. Charlie’s report is somewhat uncharacteristically informative. He has done his homework on this one. All said it’s quite impressive. Obviously, NVDA has indeed dropped the ball, to the tune of a 40% failure rate.
In 20-20 retrospect, INTC’s Prescott was a wakeup call which was clearly addressed a few years back.
NVDA wasn’t paying attention.
http://www.theinquirer.net/gb/inquirer/news/2008/09/01/why-nvidia-chips-defective
SPARKS
Anyone see this? NVIDIA and ATI/AMD are being sued for collusion and price fixing. At first I was going to dismiss it, as it seems to have been filed by a couple of disgruntled consumers, and that's usually a quick ticket to a case that gets tossed early. But there are memos and/or emails between top guys at NVIDIA and ATI that imply that they were considering fixing price levels, particularly since they felt they had the market to themselves.
Larrabee changed the latter part of that plan, I suppose. I am wondering though, if there is anything to this suit or if it'll be thrown out before it gets very far. I don't know how much evidence they have or how the burden of proof applies here (civil courts can be different than criminal court in this regard).
From the article, there is certainly enough evidence to say they discussed price fixing, or atleast a price minimum, but a few emails between upper management isn't proof that it actually occurred. However, if hard proof is found, then the upper managment can certainly face jail time over it. Several memory maker executives did over DRAM price fixing.
He’s calling them “bump cracks” which you mentioned earlier.
This isn't quite the same thing I was talking about, but it is analogous.
I was talking about solder cracking between the board and the component. If Charlie is right, the NVIDIA parts are actually failing inside the package.
Now I'm not a packaging engineer, but I know that reliability testing should have caught something like this. The shop I worked for provided boards for EMC. They make their living on selling high reliability for a premium, so I picked up some peripheral idea of how it is done.
Basically, you put the board in an environmental chamber at elevated temperature and humidity and turn it on and off over and over. Cycle testing for EMC's boards could exceed 8 weeks. If NVIDIA did comparable testing, they must have known about this problem before they put the parts out on the market.
“Surprisingly, but the 45nm process technology will not allow AMD to reduce power consumption of its quad-core AMD Phenom X4 process considerably and the new chips are projected to have 125W”/
I think people miss something when they look at Shanghai. The part has at least 3x the cache of Barcelona, and I'm sure there are other improvements in there. I suspect it is those improvements and changes that have kept Shanghai from improving power levels.
It is also important to remember that TDP doesn't reflect real power usage very well. Even overclocked, the processor isn't running all out 100% of the time. It will be interesting to see what the real world power consumption on these look like. I think it will be better than Barcelona, personally. Not great, but better.
Since AMD is probably locked into the same gate thickness as they have on 65nm, I don't see how they would get much improvement on transistor speed. Basically, I think they are locked in to what they have until they go with HK/MG.
It will be interesting to see what the real world power consumption on these look like. I think it will be better than Barcelona, personally. Not great, but better.
The active power should get better as you will likely see lower Vcore, which is driven by the lower Vt for the 45nm process. This is usually significant but it will be offset to some extent by the increase in overall leakage.
In terms of speed, the process will be marginally better as there are some non gate oxide related improvements; also the lower active power should allow them to squeeze a bin (or maybe two) of speed out. I'm working under the assumption that the chip speeds are limited by TDP considerations on 65nm, which I think is a fair assumption. Of course the extra 4MB of cache is going to eat into any power gains. It's hard to say how much of a tradeoff this is - the extra cache is a lot of tranistors (4Milx8x6) , but generally the cache is a "slower" (higher Vt, less leakage) transistor so it doesn't leak/consume as much power per transistor as some of the critical speedpath circuitry.
I would think AMD should be able to get to 2.8-3.0GHz on 45nm (without highK). The 0MB L3 parts may go another bin up - though I'm not sure AMD would do this as the extra speed is just being traded off with the performance hit of the lower amount of cache (i.e. a 2.6GHz 6MB L3 part may be as fast or faster than a 2.8GHz 0MB L3 part)
!!!!!!!HOO RA!!!!!!!!
SPARKS
“but generally the cache is a "slower" (higher Vt, less leakage) transistor so it doesn't leak/consume as much power per transistor as some of the critical speedpath circuitry.”
Interesting, in Charlie’s excellently done 3 part NVDA report he had some IR shots of the GT200 core itself. The photos revealed just how localized and varied the heat was across the die.
True, NVDA chips are as large as subway mosaic tiles, and they are GPU’s, but I’m quite certain these localized hot spots share a commonality with all large scale IC’s.
Vt = threshold voltage
Wouldn’t going to 45nM (smaller depleted region) increase subtheshold leakage? Therefore, they shrink it, add some cache, lower the voltage, and the chip consumes less power overall, yet when they raise the clocks, they still faced with same leakage issues, if not worse locally (at the junction) then they did at 65nM. They got a net gain, sure, then lost it for 2 or 3 hundred MHz (with the additional cache).
I’ve seen reviews where the thermals of the Pheromone rise exponentially for every 100 MHz increase.
Of course I’m guessing, but this could be just a test bed, AMD’s version of a “toc” for their new “tic” in 2009, at 45nM.
SPARKS
“ I would think AMD should be able to get to 2.8-3.0GHz on 45nm (without highK).”
No sooner said then done. Nice call.
http://www.theinquirer.net/gb/inquirer/news/2008/09/04/amd-desktop-roadmap-leaked
SPARKS
This, too.
http://www.chw.net/images/breves/200809/1220492096_AMD_Roadmap_Logo.JPG
SPARKS
Wouldn’t going to 45nM (smaller depleted region) increase subtheshold leakage?
Yes and this is one of the challenges of scaling to smaller nodes. Until ~90nm (maybe 65nm) technology node, the subthreshold leakage (leakage from source to drain when the transistor is theoretically off) was the largest component bigger the gate oxide leakage. Around 90-65nm the gate leakage was/is on par with subthreshold, though with introduction of high K, subthreshold leakage again may be #1 (not sure).
One way around this is what is known as a dual Vt or even tri Vt process. What this means is that the transistors you need to be fast you target a low Vt which allows the transistor to be switched faster - the cost of doing this is increased leakage. For the parts of the circuit that don't need to be as fast (cache is a good example), you can set a higher Vt device which means lower leakage. Both Intel and IBM have been doing dual Vt processes for at least 2-3 generations now (in fact I think IBM may be doing a tri Vt process - the main cost of doing this is additional lithography and implant steps (2 additional sets of each as you have to do this for both NMOS and PMOS)
When you look at the heat maps of a chip you are seeing 2 main effects - one is density and the other is speed. The high speed and/or higher density will obviously run hotter and typically this is the "logic" portion of the chip.
Sparks - if you were wondering how power goes down if lower Vt increases leakage - this is because the lower Vt also lowers ACTIVE power substantially. However on and off power consumption is now almost equal so the power benefit of lower Vt's is starting to lessen.
I’ve seen reviews where the thermals of the Pheromone rise exponentially for every 100 MHz increase.
This will happen on most chips when you get near the cliff. I'm not sure where this is on Intel's chips as I'm not a serious overclocker - but it is there as well (just happens to be at a higher clock these days). I'm sure if Intel introduced a 3.4 or 3.6GHz chip you would start seeing a similar effect.
This is the main problem with the fanboy extrapolation and use of TDP's to project bins. - when you get near the top bins you no longer can do a simple linear extrapolation. For example if you get 3 speed bins going from 65Watt to the 95 Watt bin, this doesn't mean you will also get 3 bins from 95W to 125W.
I ran across this little tidbit earlier this week. I'll be interested to see just how much detail Intel plans to release.
“This will happen on most chips when you get near the cliff. I'm not sure where this is on Intel's chips as I'm not a serious overclocker”
I can’t speak for, (please excuse the arrogance) ‘lesser’ Extreme Chips, but I can speak for Top Dog QX9770. And, I can tell you exactly where.
The cliff is a little over ~ 4.2 gig, normal air cooling. If you start pushing the quad even 100 MHz over, the usual boots and reboots, coupled with throttling, makes for a bad day. INTC (wisely) put a nice juicy thermal limiter to save your bacon from Extreme frying. The Voltage is what I consider a little uncomfortable at 1.4375 V. As you know (and most don’t), this NOT actual core voltage. This voltage is properly called CPU DC Line.
Actual PLL core voltage swings (1.175V max) can be addressed, both low and high for each die. Personally, I wouldn’t touch these BIOS parameters without Mark Bohr sitting next to me, with 4 Fluke 45’s, seated between us, monitoring each dies voltage trace/swing. (Picture that in my half-assed lab, now that would be a Kodak moment!)
In any case, speaking for the Top Dog, ANYTHING below 4 Gig is cake, rookie territory. Anything above 4.2 Gig the thermals go through the roof and serious tweaking is definitely in order, not to mention cooling. (An interesting note here, BTW, QX9650’s comfort zone is ~ 3.6 to 3.8, according to MaximumPC.)
With QX9770, INTC picked up an extra 200 to 300 MHz, for kicks and giggles. The ‘Official’ thermal parameters went up a miserable 5 or 10 watts. Big deal, Pheromone nearly had to reinvent the wheel to get the same.
There’s more. GIANT has found his C2D VERY overclockable as the die works much better solo at 4.3 ~ 4.5 Gig, amazingly enough. Obviously, sharing voltage and heat ain’t a good thing.
All said, this is an extremely high cliff considering INTC offered up 3.2 Gig for the SKU. Did I say sandbagging?
“ Sparks - if you were wondering how power goes down if lower Vt increases leakage - this is because the lower Vt also lowers ACTIVE power substantially. However on and off power consumption is now almost equal so the power benefit of lower Vt's is starting to lessen.”
Quite right, I was. I’ve got it now, you lower the voltage you narrow the band. Narrow the band enough (with this leakage) it doesn’t mater if you’re off or on, you’re still bleeding, and thermals only get worse when you’re on.
Got it, as I suspected, they really are on the edge.
Thanks!
SPARKS
"This will happen on most chips when you get near the cliff. I'm not sure where this is on Intel's chips as I'm not a serious overclocker - but it is there as well (just happens to be at a higher clock these days). I'm sure if Intel introduced a 3.4 or 3.6GHz chip you would start seeing a similar effect."
Typically, dynamic power is a linear function of frequency, given by a very simple (and easy to derive) expression P=C*V^2*F (C=capacitance of the load, V is voltage, and F is frequency). You can see why companies would like to keep voltage low as it is quadratic to power. Mother nature, though, is never kind and to get clock frequency higher requires higher voltage (a catch 22), in fact, frequency goes linear with voltage, so you will often see people show the voltage dependency on power as approximately cubic in voltage.
Nonetheless, the dominant leakage mechanism up to and including 90 nm was sub-threshold leakage, i.e. the residual leakage from the transistor in the off state (think of a leaky faucet or valve). After 90 nm, gate tunneling leakage became the dominant leakage mechanism (hence the move to different gate materials in order to 'thicken' them physically but mimic ultra thin electrically).
The dynamic power term, stated above, does not account for either leakage (wasted power). However, gate leakage only occurs when the transistor is in the 'on' state as voltage is applied to the gate electrode (you can't get leakage across the gate if the voltage is 0 or off). So the reason it rises exponentially is because gate leakage is exponential with voltage and as frequency goes higher the time average 'on' time is higher (as well as higher voltage to achieve this frequency). I.e. increase voltage and you get exponential in leakage not quadratic as expected, and if you hold voltage fixed, and increase frequency (while being at least stable), you still get an exponential increase in gate leakage.
For Intel, you will indeed see similar behavior -- but Intels process technology at 65 nm is superior to AMD's 65 nm technology (based on what has been published at IEDM for each companies tech, it could be different by now but I doubt it based on current Phenom limitations and behavior of the most recent G step CPUs). However, you can see the exponential turn on even in Intel's 45 nm process, but not until much higher voltages and slightly higher clock frequencies than you are stating ... Anandtech did a nice article on this (probably on of their bests to date): http://www.anandtech.com/cpuchipsets/intel/showdoc.aspx?i=3184&p=3 here you can see a pretty decent linear relationship up to about 3.8-4.0 GHz, then it turns over into a much more rapid (exponential) increase.
AMD's 45 nm process will suffer from the same limitations as their 65 nm process ... I.e. I would expect them to struggle to get up to 3.0 GHz (which is probably doable at 45 nm), but I don't expect much more than that at their target thermals. This will change if they proceed to a new gate material like Intel has done.
Jack
AMD's 45 nm process will suffer from the same limitations as their 65 nm process .... This will change if they proceed to a new gate material like Intel has done.
Unfortunately, I don't see that happening any time soon. The latest from IBM was that they were introducing HK/MG at 32nm. (As opposed to "late 45nm", which was just plain bogus.) They went on to say that they expected many of their development partners to "skip" the 45nm node entirely to take advantage of the benefits HK/MG brings.
To me, that all paints a picture of AMD being left holding the bag at 45nm. Given their current financial straits, I don't see how they finance 32nm any time soon.
To me, that all paints a picture of AMD being left holding the bag at 45nm.
I've said it in the past - the only way I see AMD doing highK/MG on 45nm is if 32nm appears very late or broken or not finance-able (if that's a word). This is not a mere CTI step which can be dropped in - you are talking about a whole new set of product certification as the gate oxide reliability has to be validated and is non-trivial, and you will likely need to re-layout the chip to take full advantage of the gate improvement, not to mention various upstream and downstream process steps that need to be tweaked/re-targeted/tuned for the new gate oxide. Recall Penryn was not a mere shrink of 65nm - yes Intel did add some instructions but I'll bet they tweaked the layout to take better advantage of the new gate oxide as well.
So in addition to significant process changes, you are likely talking a whole new round of tapeouts - this would seem to be a rather large resource investment (engineering, silicon, money, fab capacity hit to do the piloting) if it is just a late 45nm thing - which would imply a short time period before moving to 32nm.
Unless 32nm is way late (which it supposedly isn't according to IBM) or will be pushed out by AMD for other reasons, I don't see the business and risk side of inserting highK/MG late in a technology node cycle - it needs to be put in up front where you can get adequate return on the risk, money, time and resources spent on the effort. The armchair process technologists (on other boards) don't seem to understand this change is more than just a matter of bringing in the new equipment and development of one or two new process steps.
LOL, how does lowering Vt making transistors go faster lower active power? Faster transistors have more drive current and push more charge. Pushing more charge burns more power. Sparks very careful of the stupid anonymous poster here as many have no clue.
4.1G overclock, damm AMD is really holding a bag, a bag of IOUs and nothing much else.
It didn't take long for Lex to come out of his hole, whenever the conversation turns interesting, Lex will be riding in to crap on the discussion.
Ortho are you BIC? If not get back to work
It's called vacation ;)
LOL, how does lowering Vt making transistors go faster lower active power?
I'll take this as a serious question...
Vt = threshold voltage, which is basically the voltage at which the transistor turns on. By lowering this voltage you can get more overdrive which is the difference between Vcore (effectively) and Vt. The higher this difference, the more "charge you can push" and the faster you can switch the transistor. Thus lowering Vt allows you to either lower Vcore (and get the same amount of overdrive/transistor speed) or keep Vcore the same and get more speed.
What you neglect is you also get speed through other means as well (strain, implant, oxide scaling, etc) so you can drop the Vcore with the Vt and still manage to get transistors faster. I think Intel has been doing this for just a few generations (sarcasm intended), last I checked. So yes lower Vt = faster transistors and you can also do this at the same time as lowering Vcore and lowering active power consumption.
Unless you disagree and think Intel has not been both making faster transistors and lowering Vcore for numerous generations?
“Another issue is the thickness of the gate oxide. A thinner oxide allows the transistor to be switched on and off faster, but it also increases leakage. The amount of leakage is also influenced by the threshold voltage of the transistor. The threshold voltage (VT) of the transistor is the voltage at which the channel conducts current between the source and the drain.
Small high-speed transistors need a lower threshold voltage (influenced by oxide thickness and doping) to
maintain the speed with which the transistor can be turned on and off via gate control, but this increases the leakage because the transistor channel can not be turned off completely.”
Hmmm, from my own independent reading, I’ve found this. Also, INTC has, over subsequent generations, has lowered voltage and speeded up transistor speed as far back as I can recall. Now they are dealing with the previously mentioned issues AND simultaneously reduced leakage with Hi-K.
Additionally, they will be shrinking the ‘Vt band’ as they have over decades. They will lower voltages further and drastically reduce leakage with both FD-SOI and Tri-Gate Transistors.
So why are you Laughing Out Loud?
It’s really history at this point.
SPARKS
Orthogonal! It's good to have you you back. I thought we lost you, too.
Let me quess. Rio, with plenty of G strings as thin as CPU traces!
SPARKS
“here you can see a pretty decent linear relationship up to about 3.8-4.0 GHz, then it turns over into a much more rapid (exponential) increase.”
“yes Intel did add some instructions but I'll bet they tweaked the layout to take better advantage of the new gate oxide as well.”
Both correct and substantiated with the tweak of QX9770 from QX9650.
Sure, same core, same cache, bla, bla, bla, as far as I’m concerned they are two very different animals. The frequency headroom and voltage tolerance is excellent, not to mention “but I'll bet they tweaked the layout”. F**Kin A bubba! That’s what I paid an extra 500 cannolies for, and I got my money’s worth.
This thing makes it easy to drop that kind of cash. Everyday justifies the cost.
SPARKS
Meyer drops the bomb:
“We’re going to go away from a captive fab model to more of a fables model for the CPU part of the business,” Meyer says. “Longer-term, it relieves us of the burden of having to shell out cash for these gigantic factories. So it will be more of a pay-as-you-go model like a traditional fables semiconductor company.”
I wonder how they will secure leading edge capacity for their CPUs. That stuff is really expensive and requires significant commitment of money to develop and use that capacity.
AMD pretty much with this has given up competing with INTEL on technology and going to be nothing more then a Via. It is like trying to be a superpower with no military industrial complex. You stand inline for the second rate weapons and Meyer will be standing inline for second rate silicon
AXEL: From the same link.
“Retailers, Meyer says, perceived the shortages as “AMD screwing our [manufacturing] customers and screwing retail,” Meyer says, “so they said let’s take a couple of steps back from AMD for a couple of quarters.”
Peh, lame excuses and sit on it and spin, this is what they bargained for, if not begged for, back in late 2006. “DELL at all cost” was the mantra. The freight train was screaming down the tracks with C2D, Barcelona pipe dreams, and volumes they couldn’t sustain for DELL and the channel, not to mention killing socket 939 with 940 that no one wanted. Heh, they screwed everyone. Bad move after sequential bad move was the order of the year; I believe they called it the “perfect storm”
And what say ye on a fat 1.2 billion from the New York State Taxpayers? (Ah, that’s Luther Forrest to all those that are uninformed.) “Ah, we’ll take the money first and see how fabless, fabless is!”, that’s after we screw the Germans with Dresden, of course. Fab 30 is mothballed, basically junk for the Ruskies, and 36 needs a revamp. Enter Arab MiddleEast Devices.
Nice try Dirk. What does the INTC license agreement say about chip outsourcing again?
Actually, they’ve got so much crap on the burners; they don’t know which pot to stir first, neither do we.
SPARKS
Actually, they’ve got so much crap on the burners; they don’t know which pot to stir first, neither do we.
Hate to break this to you, but I think they gotta go with the NY Fab first. After the suckers, ummm politicians, decided to reach in everyone's pockets to the tune of 1.2Bil (plus the hidden infrastructure costs), AMD is probably better off starting from scratch with a 25-30% off coupon courtesy of NY.
Yeah a F30 revamp centralizes operations, makes it easier to share support organizations, but 1.2Bil is 1.2Bil and that is probably the only real bargaining chip that incentives a joint venture partner.... otherwise you are asking someone to take on billions in debt and a 65nm/45nm fab that is currently operating at a loss and a shell of a fab that needs to be ramped. Where do I sign up to donate money to that!?!?
Even with AMD's fab financial model, a $1.2Bil off coupon has to give even them a chance of turning the operation for that fab into the black.
Say a fab is 4Bil for the fab and equipment (probably a bit on the low side now depending on size, but it makes the math easier) - 1.2Bil is ~30% off. If the fab and equipment costs is ~60% of the total wafer production cost (a bit of a guestimate), you are now talking about production cost savings of 30% x 60% or 18% - that is rather significant.
This is a bit oversimplified but it gives an idea of how much that sort of subsidy can mean to a company operating on the edge of profit/loss. (also keep in mind that this doesn't mean NY will be operating this much lower than Dresden as I think AMD still gets some gov't payola in Germany too)
"Where do I sign up to donate money to that!?!?"
I'm surprized at you, a man of your intellect! That's easy! Albany, New York via a signature on your NYS Income Tax W-2.
I do it every year. It comes out of my paycheck, every week, just smooth things along.
After all, when they do go though will the deal, they might have to justify their "investment" with additional subsidies in the future.
"A little give and take"
"The New York Tumes"
"The Daily News"
"Whoa Ohhh"
SPARKS
Build a multi-billion fab with handouts so you can absorb billions more in depreciation running an inferior process a year and half later then INTEL, make perfect sense. Go Dick Go! Get AMD more into debt because you got some free handouts from NY taxpayers.
"A little give and take"
"The New York Tumes"
"The Daily News"
"Whoa Ohhh"
You just sound like "An Angry Young Man" to me!
And you forget my money no longer goes to Spitzers' consulting firm or was he consulting some firm... ahh never mind... there's a joke in there somewhere! New York is about as corrupt as it gets and is a place where the carpet-baggers come so they can eventually run for higher office - can't say I miss it much.
In other news, as part of Asset Smart AMD will teach the Arabs how to implement LEAN and SMART production techniques into the fabs they will be selling them.
Well folks, while the 2005 through 2008 focus was on the “evil empire, “monopolist INTC”, the rest of the industry was running amuck. We at this site have focused mainly on AMD’s shirt and spin failures. However, in other news today, some big players have come above the radar, and these will be juicy targets.
First there’s Yahoo. It is now a target of a DOJ investigation.
http://www.smh.com.au/news/biztech/googleyahoo-deal-facing-new-scrutiny/2008/09/10/1220857598231.html
Google and Yahoo together you think? Sure, when pigs whistle.
Then the abomination called Nvidiot has had a suit filed against it by a major share holder for withholding information concerning critical chip failures while pumping up the “whoop ass” strategy.
http://www.free-press-release.com/news/200809/1221080651.html
Hmmm, I wonder how many others are going to jump on this train. I think more than two is called a class action suit, (caution: extreme sarcasm here).
Ya got to these high powered arrogant bastard corporate Exec’s. They really know how to show their shareholders how much they care, and who’s the boss.
Wrector Ruinz and his minions are in good company.
BTW: If you think AMD’s monumental disasters were a major daytime drama for the past few years, this NVDA suit will be a huge block buster. Oh, I can’t wait for the discovery portion of the suit. I’m sure many of NVDA’s previously strong-armed ‘partners’ will be more than willing to share some juicy tidbits with the press and the DOJ. As we say here in the Big Apple, “payback is a bitch.”
Did I say stay don’t even think about catching this piano?
Intel doesn’t look so evil anymore, eh?
SPARKS
Oh, yes, about INTC? If you were fortunate enough to be a shareholder of record, they just paid a 14 cent quarterly dividend.
http://news.moneycentral.msn.com/ticker/article.aspx?Feed=AP&Date=20080910&ID=9125653&Symbol=INTC
ATOM
QX9770
CORE i7
Over a half a buck a year dividend in the worst of times.
“Man that Sparks is a real Intel fanboi”
Wait, there's NVDA, ah, no.
Hold on. There’s Lehman Brothers, no.
AMD! I don’t think so.
Wachovia! Yeah, right.
Fo Mo Co Ha, ha, ha.
BOA, nah.
Yeah, I’m a INTC fanboi, no doubt, but that
"SPARKS ain't stupid!"
SPARKS
By the way, seriously I might add, if any you on this site are serious NVDA shareholders of record, and you lost a great deal of money, you might consider filling out one of these:
Personally, I would speak to my Attorney if the amount was substantial.
http://www.lawssb.com/docs/NVDA_Certification.pdf
Here is another link for additional news regarding NVDA nearly $30 a share drop:
http://www.xbitlabs.com/news/video/display/20080910112030_Nvidia_Accused_of_Fraud_by_Investor_for_Not_Disclosing_Information_Regarding_High_Failure_Rates.html
SPARKS
Intel doesn’t look so evil anymore, eh?
That depends on who you are. I took a very rare stroll over to AMDZone just the other day, and I can assure you there are folks over there that will die convinced that INTC is the root of all evil. Every success Intel enjoys is only further proof that they are evil.
While there, I couldn't resist browsing through their thread on there take on AMD's forthcoming answer to the Atom processor. Predictably, Atom is a worthless piece of junk that no sensible person would buy. If it weren't for Intel's monopoly position, nobody would touch it. Yada, yada yada.
The thing that I thought was interesting was the way the only metric that really matters was casually dismissed. There is HUGE demand for this thing. And it isn't the demand for Atom that matters, it is the demand for this form factor. But since AMD isn't a player in this space, it is all rubbish. At least until their new "UVD" part comes out, then I'm sure it will be a real AMD coup in their eyes.
Still, it might be worth taking a look over there since Das Tirpitz seemed to have some interesting insight into where this whole thing is heading (I think everyone here is all too aware of my position). Of course the AMD faithful are all up in arms with his assessment.
I do have to wonder what is going on with Intel's Poulsbo chipset for the Atom. That chipset brings the total solution down to 5W or less compared to the 25ish W of the 945 chipset derivative. You would think that the Poulsbo chipset would be the first choice to put Atom in a class that no one else could touch. But that doesn't seem to be the case.
Poulsbo seems to be virtually non-existent at this point. I have to wonder if Intel is guilty of a paper launch on this chipset.
What is good for INTEL is good for computing.
The better the CPUs' they can make, the more desirable they make and market them the more they will sell. The more they sell the more money they make, the more money they make, the more profits they earn. With more profits then can invest in even more innovation.
Yup a big profitable INTEL is very very bad. Who wants faster computers, cheaper computers, ever more pervasive computing.
Nah, its all foul play in AMD fanbois eyes. INTEL just wants to sell their own CPUs exclusively to make a profit for profits sake. Better in the AMD fanboi's eye to cut INTEL down at the knees and stifle them and penalize them for their success.
Believe it or not there are many cases in business where the people did better when there was a monopoly then when there wasn't. Only when monopolies collude to stifle change, competition and innovation is the public harmed.
"Did I say stay don’t even think about catching this piano?"
NVDA has reached 3 year lows on an "overweight" downgrade, $12 from a previous $15 target.
http://news.moneycentral.msn.com/ticker/article.aspx?Feed=AP&Date=20080911&ID=9132498&Symbol=NVDA
Please note, this was on the heels of an entire market boost today. (Although this boost may be a "dead cat bounce".) NVDA shed over a half a buck!
However, speculators be warned. If yesterdays class action suit has any teeth, NVDA will give a new meaning to "opening up a can of whoop ass".
NVDA is at 10.52
SPARKS
Fella’s , does anyone know what’s up with this or it implications?
http://www.digitimes.com/bits_chips/a20080909PD224.html
INTC can’t pump out enough of the SOB’s (Atom shortages), conversely, TSMC has a 300mm slowdown?!?!
Does anyone have a clue?
SPARKS
INTC can’t pump out enough of the SOB’s (Atom shortages), conversely, TSMC has a 300mm slowdown?!?!
First of all, Intel doesn't have a manufacturing capacity issue with Atom. What they have is a testing capacity problem.
Think about it, you get a 4x boost in the number of chips per wafer with the Atom and now you have to test all those chips individually. You combine that with better than expected demand and you have a shortage of Atom chips in the market place.
As to TSMC, I think there are 2 things at work.
First, semiconductor demand in general is down. The exception to this is CPUs. Since TSMC isn't making CPUs, their demand is down.
Second, I read somewhere just today one of the big issues the foundries face right now is that their customers don't have enough visibility into long term market demand. So they are playing conservative and slowing their rate of expansion.
I remember rambling on about speech recognition a while back. I think I suggested it might fly at 22nm.
Paul Otellini must be reading this blog, because he seems to be on the same page. ;-) He is saying that current processors don't have enough power to make this fly yet, but it is one of the goals that Intel is working towards. You can listen to his comments here.
ITK: Second, I read somewhere just today one of the big issues the foundries face right now is that their customers don't have enough visibility into long term market demand. So they are playing conservative and slowing their rate of expansion.
This is dead on (and I believe was TSMC)- with overall solid semi growth in the past this was not much of an issue as foundries could always find some one else to buy whatever excess capacity they had.
With growth slowing down it is risk to build in excess capacity as if it is not used it starts crunching the margins and profitability. The migration to 300mm also makes this worse - the capital contribution to the finished wafer cost has gone up so if you have excess capacity you are effectively taking a bigger financial hit for underutilization.
In other words if you don't produce a wafer you save a good chunk of the variable costs like Silicon and chemicals and consumables but the capital depreciates whether it is running or not. On 300mm the capital portion of the wafer cost has gone up so this hammers under-utilization more so then 200mm.
However, speculators be warned. If yesterdays class action suit has any teeth, NVDA will give a new meaning to "opening up a can of whoop ass".
Sparks - even if the class action lawsuit does indeed have teeth you are still talking way out in the future (depositions, the usual delay tactics, trial, potential appeals, etc...) - I can't see how this would have any additional impact on Nvidia for the short term(next 6-12months) unless if some damning email or something leaks out.
That said Nvidia is a crapshoot from an investor perspective - you still have unknown one time costs from the bump cracking issues (though Charlie's BK scenario seems a bit far fectched). You also now have huge competitive and pricing pressure from AMD which looks like will continue for the foreseeable future. Finally it looks like you will have continued loss of chipset revenue (though this probably has a minor impact to the stock price)
BTW - you're upper 20's INTC price is looking a bit questionable - as a stockholder I'm a bit confused by the recent price action (it is not clear any sort of fundamentals are driving the recent stock price moves)
anon: "You also now have huge competitive and pricing pressure from AMD which looks like will continue for the foreseeable future."
I wonder how AMD fans feel about that. :)
“as a stockholder I'm a bit confused by the recent price action (it is not clear any sort of fundamentals are driving the recent stock price moves)”
You’re in good company and you how I value your opinion.
My Wall Street buddy (Gold) who I ride the LIRR with, has a different take, and he is scaring me. His words were, “We are one more bank failure from a total bank meltdown”. Also, he did mention Lehman Brothers, and it ain’t pretty. They are scrambling.
Actually, INTC is holding up VERY well considering the current market conditions, according to him, and I tend agree. “Count your blessings”, said he.
Uhh?
That said, I’ve noticed every time INTC goes below 20 (today’s 19.75 for example) there is a flurry of buying that sends it right back. That’s two days in a row. It closed at 20.16. According to him this is GREAT as they, obviously, have great fundamentals. But we (you and I) knew that.
I still maintain, and I am biased, that INTC will take off if we can get out from under this banking catastrophe any time soon. That’s a big ‘if’, and there’s no light at the end of the tunnel, yet.
Time to buy?
Nvidiot? It ain’t no crap shoot, it’s a lead balloon. There isn’t enough hot air in Jing Hung Sungs lungs to float this pig. Even he mentioned, rather casually, that the margins aren’t there because of competitive pricing pressures.
YA THINK???
Let’s see where the market share numbers are at the 3Q call. Expect another hit.
With die half the size of Saltines they can no longer command a luxurious $1000 for a second rate product, whoops! And this is coming from yours truly who doesn’t mind spending a thousand buck on a high end product. Further, at these prices I’d buy two, in a New York minute. Then again I can’t, there’s no SLI on my X48, perfect 20/200 myopia.
I’d love to wave $1200 in front of his face and say, “See, a$$hole, you could’ve had this.”
NVIDIOT
SPARKS
HA!
Do I smell an INTC COUNTER SUIT???
http://www.theinquirer.net/gb/inquirer/news/2008/09/12/intel-man-accused-secrets-amd
SPARKS
"Do I smell an INTC COUNTER SUIT???"
Probably not - they'll just make sure the lawmen get him.
As for AMD, they'll probably have to cut this guy loose to avoid any perception that they could have been involved or benefited. Though indications are they had no knowledge of this as most of the secret documents were found in his home.
That said apparently he was being logged when he was doing this so it will be extremely hard for him to argue negligence and no intent on this. I wonder what the going rate for intellectual property theft is?
“I wonder what the going rate for intellectual property theft is?’
A dinner date with one of Henri Richards’ former “Administrative Assistants” (you supply the silk ties), a cubicle decorated with a color of your choice, a free Triple Cripple, an AMD ‘Outstanding Innovation Award’, one autographed picture of Wrector, and a box of ‘Milano’ cookies.
‘Desperado,---- why don’t you come your senses……’
The Imitator
SPARKS
I was recently looking to buy a basic laptop for my wife to access Internet & email. We initially thought of buying one of the Atom netbooks. We went around the shops looking for one. Our observations were :
- Screens are too small if you want to use it for longer periods, especially if you are going to be reading news articles and stuff.
- They look tacky, not half as neat as most of the laptops. If these are devices which people are carrying around, there needs to some effort in improving the looks.
- At $299 for a basic one and upto $450 for those with HDD etc, prices are too high for what you get in the package (features+looks). You wouldn't be paying too much more for a low end laptop which will have much more functionality.
After considering all the above, we ended up buying a full featured Gateway laptop ($800) . I think we appreciate what we bought, much more than what we would have without having looked at the netbooks.
I think the netbooks need to improve by a lot, if they have to see a sustained market in the mature markets. With respect to developing markets, there have been other cheap laptop initiatives but they haven't been so successful. Let us see how far this will go. In any case, Intel will gain long term as this will spur demand for full featured machines.
I am surprised that there are so many Atom-based notebooks out there. That didn't seem to be the area that they were aiming for, I thought they were looking towards stuff like cell phones and PDAs. I assumed it was a first-gen part that wasn't going to be very successful in terms of sales. But there are a lot of Atom notebooks out there, it seems.
I'm wondering if Larrabee will shoot out of the gate as well, even though it also impresses me as a technology that is on the ground floor.
Tonus said...
I am surprised that there are so many Atom-based notebooks out there. That didn't seem to be the area that they were aiming for, I thought they were looking towards stuff like cell phones and PDAs. I assumed it was a first-gen part that wasn't going to be very successful in terms of sales. But there are a lot of Atom notebooks out there, it seems.
From http://www.dailytech.com/Report+Lowend+Portables +Help+Global+PC+Sales/article12929.htm
"According to a new report from IDC called IDC's Worldwide Quarterly PC Tracker, low-cost portable systems are allowing the PC market to resist economic pressures. Despite the current economic woes in many areas, PC shipments are still growing."
"Shipments are especially growing in the low-end portable market reports IDC. The low-end portable market is expected to continue driving growth for the next few years. Worldwide PC shipments are expected to grow by 15.7% in 2008 to 311 million units."
Offsetting these financial expert reports, we have the illustrious Ah-Ben-Stoopid over on AMDZone claiming that Atom is a failure, simply because it is not as fast as an AMD Opteron. Of course because of his vast influence and expert standing within the entire semiconductor industry, Intel will immediately cease manufacturing the Atom and instead license Opterons from AMD...
NOT! :)
Since you guys are the subject of ATOM. Here’s an interesting take on AMD’s answer to it. Ed at overclocker is calling AMD’s release “a stunt”. Huh?
Anyway, low muscle, low power chips are about as interesting to me as taking out the garbage. But, I though you might find Ed’s take interesting.
http://www.overclockers.com/index.php?option=com_content&view=article&id=4220:amd-lets-underclock&catid=57:processors&Itemid=4263
SPARKS
Now this is more like it! Fuddy has got a bunch of stuff on Nehalem, especially this monster!
http://www.fudzilla.com/index.php?option=com_content&task=view&id=9452&Itemid=1
Plus, he’s got something on a 32nM “tick” in late 2009. Man, are they executing or what? I just got done breaking in a 45nM gorilla!
SPARKS
Doc, ITK, it’s seem your estimate of how long it would take CRAY to capitalize on its deal with INTC has come sooner, much sooner, than you originally speculated. True, these aren’t full blown Computational Fluid Dynamic machines. They still are classed as HPC machines, however.
That was quick, 60 G’s, and cheap.
http://investors.cray.com/phoenix.zhtml?c=98390&p=irol-newsArticle&ID=1197689
Interesting, the link calls it ‘Phoenix’, as in ‘rising from the ashes’.
Hmmm.
SPARKS
Ed at overclocker is calling AMD’s release “a stunt”. Huh?
Stunt may be a bit harsh, but underclocked, LARGE DIE, dual core (or single cores on 90nm) is not sustainable. The performance might be there but cost will be an issue and only get worse moving forward.
What happens with future node transitions? Is AMD going to shrink K8 to 45nm to keep selling these things? While straightforward, this is still time, money and another design to support as they move on to K10.
The strategic problem is AMD is already trying to eat the low end mobile market. So while netbooks with atoms may be popular right now and AMD may feel a need to compete, in reality they are probably also doing damage to their current mobile biz and just bring their mobile ASP's down.
I've seen comments on AMDzone saying well you can't look at die size because you are comparing 90nm and 65nm AMD CPU's with 45nm - but these clowns are under the silly assumption that AMD is going to keep shrinking these soon to be EOL designs. Do people really see AMD doing a single core 45nm K8? Are they going to re-spin the design AGAIN if/when they go to highK on late 45nm/32nm?
That said if they have the capacity and are not operating at a loss on these things, it may make sense - but this is assuming this is not stealing some sales from slightly higher clocker (and better margin) AMD mobiles.
anon: "Stunt may be a bit harsh, but underclocked, LARGE DIE, dual core (or single cores on 90nm) is not sustainable. The performance might be there but cost will be an issue and only get worse moving forward."
I think that is what Ed was referring to as a stunt; a move that shores up a short term issue without addressing the long term. If AMD can produce a part that competes well with Atom on performance while lagging on power, it might be a mild success. But it's not a long term solution, and if AMD doesn't have a true Atom-smasher (har har) ready to go before very long, they'll be right back where they started... or somewhat worse.
I still get the feeling that Atom's sales success is something of a cherry on top for Intel. It's an early iteration of a technology that they plan to continue to develop and improve. That should give the follow-up technologies an even bigger PR boost when they arrive, assuming they hit their speed and efficiency targets.
And if they do, by that time what will AMD do? Introduce a "new" line of underclocked Phenoms with only one working core?
Well since Intel has blown away the high end desktop market, and soon will do the same to the server market, the only thing left to speculation is the ATOM.
This too, apparently, shall pass.
VIA’s Nano has failed to impress.
(Tonus, well said, ATOM is shaping up to be a very big “cherry”, indeed.)
SPARKS
http://www.fudzilla.com/index.php?option=com_content&task=view&id=9406&Itemid=1
Sparks... Let's not put too much stock into Fuddie.
Is it an even or odd day? The reason I ask , is that I'm just curious if Fuddie will right an article that a integrated Nehalem/IGP is on target for 2009 or late and for 2010.... how many times is he going to report on it (and then eventually claim he was right all along)?
Atom's success/failure at this point is probably 99% related to Intel. The netbook market will likely be driven on Intel's next IGP to be coupled with Atom and eventual SOC design. I have to think Intel paired it with 945 half thinking - 'well this is just a trial balloon and if it really takes off we'll get the 2nd iteration right.' Maybe a bit arrogant, but this is a new space and it is not like they are up against much at this point in time.
I saw a funny news headline. Figured I should share the laugh.
http://www.telegraph.co.uk/news/uknews/2983652/Baroness-Warnock-Dementia-sufferers-may-have-a-duty-to-die.html?source=EMC-new_19092008
Baroness Warnock: Dementia sufferers may have a 'duty to die'
Ok, sorry for the interruption. Carry on.
And as if on cue Fuddie claims there was a 'motherboard glitch' on the Tegra testing!
And did I say right or write? Good grief!
I thought some of you might be interested in the performance of QX9770 after 5-6 months of abuse @ 3.84 GHz @ 1.41V.
Since it’s nice and cool in the Northeast these days, I thought I might take the Baby out for a 4.28 GHz spin, to test it’s long term viability. Guru’s overvolting, atomic tunneling, trace degradation, insulation layer/junction breakdown mentioned earlier, made me a bit of a sissy.
Since Core i7 965’s release is so close, I felt like throwing caution to the wind (What better excuse to replace this bad boy.) Despite Guru’s arch conservative, cautionary, scientific, methodical approach towards process technology, I said f**k it and opened the windows!
How’s that for technological prowess, eh?
QX9770 @ 1.45V (ouch), Multiplier 9.5, 4.275 GHz
Asus P5E3 Pre. Bios v 0505
FSB 1800 (strapped)
FSB Freq. 450
Northbridge Volt @ 1.30
SuperTalent DDR3 1800 @ 2.1 volts, 7-7-7-21 2T timings
DRAM Freq. 900.1 MHz x 2
2-150Gig Raptors, Raid 0, Native; X48
PC Power and Cooling 1KW. (Hoo ya!)
XP PRO SP3
Super Pi
Real memory =2146447360
Available real memory =1571115008
Allocated memory = 8388648
+ 000h 00m 00s [ 16K]
+ 000h 00m 00s [ 32K]
+ 000h 00m 00s [ 64K]
+ 000h 00m 01s [ 128K]
+ 000h 00m 02s [ 256K]
+ 000h 00m 04s [ 512K]
+ 000h 00m 10s [ 1M]
+ 000h 00m 25s [ 2M]
+ 000h 01m 01s [ 4M]
+ 000h 02m 24s [ 8M]
- Not calculated. [ 16M]
+ 000h 11m 49s [ 32M]
Sandra
Dhrystone ALU 66315MIPS
Whetstone iSSE3 61239MFLOPS
Multimedia Int 54547it/sMultimedia Float 25521fit/s
Cache and Memory Bandwidth 65.1 GB/s
Int Buff’d iSSE2 Memory Bandwidth 9.76GB/s
Float Buff’d iSSE2 Memory Bandwidth 9.73GB/s
Memory Latency 55ns
(So much for FSB bottlenecks)
(So much for DDR3 gloom and doom)
Orthogonal, X48 is bullet proof, the chipset reminds me of the 850E's stability.
Hard Drive(s)
Max 123.5
Average Read 114Mb/s
Access Time 8.7ms
Burst 115.7Mb/s
Well, I am very happy to report after 6 months of overclocking @ 3.87 Gig at 1.41 V, there has been no degradation in performance or stability @ 4.28------------If you’re not paying attention, that’s a GIG (!) over stock!
(See Guru, not so bad, ya me pissing my pants for months)
ALL ON AIR! Zalman CNPS9700 LED
Not one crash all day, playing/ running everything.
Unquestionably the absolute best CPU I have EVER owned.
QX9770 my all time favorite.
Barcelona never had a chance.
SPARKS
Anyway, low muscle, low power chips are about as interesting to me as taking out the garbage.
Looking at where Atom is today, I'd have to agree with you. It is the potential that I find exciting.
What can I say, I grew up with Star Trek. The idea that I can talk to my computer and have it run a search on the internet and have it give me the answers is kind of cool.
Atom is quite a ways from that at this point, but that is where I see it going.
I also found this item on mid's from TI. It has some interesting comparisons to "x86" (read Atom) in it.
I found their representation of the current state of Atom to be fairly accurate, but as with any piece of marketing, it is somewhat biased. I also have some issues with their view of the future.
So here is where I disagree with them.
First, they say Atom was not intended to target the PC market. They also make a big diea of the chipset’s manufacturing node and power. Intel knew the chipset was an issue. That is why Intel was talking up Moorestown before Atom (Silverthorne) was even released. So TI is correct about the current generation, but Intel had recognized the problem and had plans to address it before the product was even released.
Why didn’t Intel wait for Moorestown, then? I can think of two reasons. First, Silverthorne is a trial balloon to find out where some of the surprises are. Second, I would speculate their marketing research said this was the time to do this. Judging by the success of the netbook market, I wouldn’t want to try and second guess them.
Then TI goes on to analyze the power consumption. They claim (correctly) that their OMAP 3 processor uses 4 to 6 times less power than Atom. That is true as far as it goes. What they are careful not to point out is that Atom runs at up to 2x the speed of their solution. Let’s be very generous and assume that you could clock their solution up to the speed of the top of the line Atom (1.6 GHz) with a linear increase in power consumption. This would bring the power difference down to about 3x. Reality is they would be closer than this.
Interestingly enough, Moorestown is targeted to cut active power use in half. So with the release of Moorestown, Intel’s solution should be very close to TI’s solution from a performance per watt standpoint. Though their top end product will still use about twice the power of TI’s solution.
TI goes on to talk about their power advantage in sleep states. Again, this is a real advantage, but Intel has plans to address it with Moorestown. In fact, the plan is to reduce idle power by 10x. So Intel’s solution may not run for a week like TI’s solution does, but it won’t need “to be recharged several times a day”. As I’ve said before, if you have to plug it in at night while you sleep, who cares? You won’t be using it while you sleep anyway.
So TI is trying to focus attention on the current offering. While I don’t have an issue with this, since I can only buy what is on the market now, I think this sends an interesting message. TI is worried about Intel’s entry in this space.
You don’t see this kind of marketing directed at Via’s nano processor or AMD’s future attempt to enter the low end space. That leads me to believe TI isn’t worried by these attempts. TI does seem to be trying to establish mind share ahead of Moorestown, though. I have to believe that they are concerned about what Intel might be able to do in the very near future.
I thought you process boys might be interested to know that the IBM peacocks are in full display mode. If I’m reading this thing right, they are contradicting most of what I have learned here, and some.
They using the same tools to process 22nM with 32nM tools, through software????
“So the solution is to use existing lithographic tools with massively parallel computation to make smaller products. Computational scaling is an attempt to model and optimize the entire process flow for each new design. Like Intel (NASDAQ: INTC), IBM is moving to high-k metal gate design, which will be used in 32nm and 22nm processors.”
Are they suggesting they can’t, or won’t, develop the tools necessary for the 22nM node?
If this is such a great idea why wasn’t this implemented on previous nodes?
(Ahh, does this mean the peacocks have given up Hi-K at 45nM? Hey, I’m no Einstein but I can read between the lines here. Shoot me, the newbe, for speculating but I think they couldn’t get Hi-K to work @ 45.)
"There really is not another significant improvement in lens or numerical aperture on the horizon," he told InternetNews.com. "Other aspects of the equation that govern scale have reached their theoretical limits. The industry has acknowledged this now."
Haven’t they said this before, like before INTC put Hi-K to work a few years back? It sounds like Deja Vous all over again.
“IBM fully intends to go to 15nm using this technique, at the very least.”
Huh (???), what about the “theoretical limits” “at the very least”?!?!
SPARKS
http://www.internetnews.com/hardware/article.php/3772476/How+Small+Can+You+Go+IBM+Heads+for+22nm.htm
ITK- Here’s some ATOM goodness for ya.
http://www.hexus.net/content/item.php?item=15522
I must say, however, I’m a bit disappointed with weight of these things. At about 2 and a half to three pounds, no biggie. The price is right however.
Frankly I rather have a full blown light weight Toshiba that would smoke these things albeit at 2 grand or better, especially if I’m going to hump around 3 Lbs all day. It’s gonna be worth my while. That’s for sure.
SPARKS
ITK-Like this one, at 1.7 Lbs!!!
http://www.toshibadirect.com/td/b2c/pdet.to?seg=HHO&poid=389712
SPARKS
"There really is not another significant improvement in lens or numerical aperture on the horizon," he told InternetNews.com. "Other aspects of the equation that govern scale have reached their theoretical limits. The industry has acknowledged this now."
kind of funny given not too long ago was it not IBM and AMD claiming a breakthrough in EUV when they produced a working test chip? (which was really not that significant as I believe it was based on a 45nm design!)
This is typical IBM research 'publish or perish' mentality and quite frankly I would not put much stock into it gets closer to actually being used in manufacturing. It also helps out with trying to strike new development deals with other IC manufacturers for the process IP as they seem 'cutting edge'.
Post a Comment