11.12.2008

Bad For Intel, Catastrophic for AMD

Intel Corp. said late Wednesday that it expects financial results for its current, fourth fiscal quarter to be "below expectations," as the technology bellwether grapples with a flagging economy.
The chip maker said it now expects fourth-quarter revenue to be $9 billion, "plus or minus $300 million," which is lower than the company's previous expectation of between $10.1 billion and $10.9 billion.


Never mind Intel, at targeted gross margins of 55%, it will do just fine. The recuperating AMD will be the worst hit. Yet again, we can expect AMD to break it's promise of returning to profitability in the second half of 2008. We all saw what happened to AMD when Intel's gross margins drops close to 50% and needs to clear inventory. It's safe to assume things won't be different.

For now you can expect AMD to make noise and show its impoverished investors (AMD @ $2.57/share) that it is taking the necessary steps. It already announced laying off 500 employees. Soon they'll be talking about 32nm and reviving Bobcat. It's about being too little too late for AMD these days.

195 comments:

Tonus said...

That's a good point. If Intel feels a squeeze and decides to lower prices to compensate, it isn't something that they'll be happy with, but it could be devastating to AMD. If they get squeezed into an even smaller pricing spread with all of those SKUs...

Anonymous said...

The magnitude of the cuts is stunning - a change in midpoint of ~1.5Bil. I think folks were expecting Intel to lower its guidance, but this is more than most expected.

Meanwhile in AMD finance world, they are using operating income to determine profitability (not net income as the rest of the world does). I believe they have one more one time revenue stream for this quarter (the CE division sale), so expect a song and dance and hidden one-time revenue streams baked into operating income and perhaps even gross margin again. Then you'll have one time charges which will obviously(?) be excluded from operating income #'s.

You'll also likely see quarter to quarter comparison excluding the Q3 one time revenue streams (tech license revenue) even though AMD put it in their operating income last quarter and even baked it into the margins.

Find the king, find the king!

InTheKnow said...

Actually, I think that Intel's guidance is everything that AMD could wish for. If Intel is taking a hit to the tune of $1.5 billion, who could blame AMD for not being profitable. This may just be the perfect excuse to cover up all of AMD's woes and managerial mis-steps.

Assuming of course that they survive the coming downturn. The move to divest their fabs couldn't have come at a better time.

Anonymous said...

AMD currently down over 8% to $2.35. Sheesh, this despite the encouraging Shanghai preliminary reviews..

SPARKS said...

“needs to clear inventory”… That’s the key element, isn’t it?

Arab Micro Devices can spew performance per watt, pump 32nM, cook the books, and play a shell game with king sized coconuts. The bottom line will be product. INTC has a way of stimulating the industry merely by dropping prices. We’ve seen it before, and we’ll see again.

This, more importantly, may seem like a good thing for some politicians looking to take credit for a recession turnaround. “Bravo, Intel!” they will cry, as INTC drops prices, stimulating sales, and simultaneously cutting AMD’s legs out from underneath them.

“Oh, what about the Scrappy Little Company? Whoops, I guess they couldn’t handle the recession, too bad, another casualty.” Besides, the Dems have bigger fish to fry, like saving the American Auto Industry. This is gospel.

Up 91 cents today, does anyone think INTC will have any trouble? Think again.

SPARKS

Anonymous said...

Up 91 cents today, does anyone think INTC will have any trouble? Think again.

I think it'll be a quarter or two before any turnaround occurs, so I wouldn't be surprised to see AMD stock near a dollar and Intel below $10, with large swings as the investors scoop up bargains and then sell on the slightest bad news.

But my magic plan is to buy 20% AMD stock and 80% INTC with my end-of-year bonus :). That way when the economy rebounds, one or the other is sure to jump up :P

SPARKS said...

"That way when the economy rebounds, one or the other is sure to jump up :P"

While you're at it, do the same with GM and Ford. Thats where I'm going.

I missed the Chrysler bailout in the 80's, not this time.

Ya can't loose, bubba.

SPARKS

Orthogonal said...

"While you're at it, do the same with GM and Ford. Thats where I'm going.

I missed the Chrysler bailout in the 80's, not this time.

Ya can't loose, bubba."

Not if it's an AIG style bailout where the gov't takes an 80% stake in the company. Shareholder's will be wiped out. Personally, I'm staying out of this market right now. It's just too volatile, cheap is relative.

Anyway, I had no idea things were looking so bad. A 15% drop in Intel's Q4 expected revenue is huge. It came as a shock. I know lots of people are talking about how bad a position AMD might be in, but they're just the beginning. I think the DRAM maker's are in an even more tenuous situation. Micron is really on the ropes, I don't think they'll survive a deep recession. Nvidia and AMD may survive, but be put back by several quarter's or more.

Intel should whether this fairly well, although I'm not expecting much on the bonus/raise front for next year. Oh well, at least I still have a job ;) I haven't heard of any changes to future schedules or roadmaps, but if things are bad enough, I wouldn't be surprised if certain things are pushed out.

Anonymous said...

Gotta echo Othroganol... the bailouts these days involve equity stakes, so while the companies may survive - the stockholder is the low man on the totem pole. You also potentially have the Gov't with a substantial say in how the company is run.

These companies are burning (I think) 1-2Bil/month... so let's say the proposed 24Bil bailout is spread evenly among the Big 3 - that buys them less than 1/2 year... before they are begging for more money. You think these ocean-liners are going to be able to turn on a dime and change course?

When something is on fire, throwing money on the fire is not going to put out the fire... sometimes you need to deprive the fire of oxygen to put it out and then rebuild. I think we need to let these things go Chapter11, which would force them to reorganize and also force the UAW (united autoworkers union, one of the most powerful unions in America) to come to the table with some honest concessions. If they know gov't money is eventually coming, why would they make any substantive concessions?

I'm not saying these stocks won't go up, but don't assume it's a can't lose scenario.

InTheKnow said...

I'm a big believer in learning from the lessons of history. As it turns our there is a rather disturbing parallel from the 1870s to our current situation.

If the parallel holds, Intel is in pretty good shape. In the 1870s it was the companies with lots of cash that prospered.

Oddly, I can see where this would work out well for AMD now that they have access to petrodollars. They may end up as minority owners when this is all done, but I can see a path to survival for them.

Anonymous said...

The AMD analyst day presentations are on their website... I looked through a few of them and it was almost comical - everything was spun and words and schedules were parsed very carefully (and in misleading ways)

One thing I noticed is there have been a few websites speculating that high K may be inserted into 45nm midway through the node, but on the manufacturing presentation it talks about 32nm SOI plans and one bullets is "introduction of high K".

Other highlights:
- NY fab - 2012 capacity expected to come online (sounds a wee bit like the schedule I laid out a while back!)
- F38 (Dresden conversion) - no capacity expected in 2009, a small amount in 2010 (looks to be ~1/4 of final capacity)
- 32nm plans are a bit vague - AMD has some crazy picture which shows 1st test chip Q1'10 and "qual" (whatever that means) at end Q2'10. Not sure if that means add some additional time for actual 32nm product (i.e H2'10)? On a separate product presentation, they show no desktop/notebook 32nm product until 2011.
- It looks like in Q3'08 tri-core and quad cores were nearly equivalent volumes for AMD. Do they have that many defective quads or are they disabling working quads? Either way that's not a good thing is it - they either have a crappy process or are taking it in the AS-...that last letter is a P! Meaning if the process is "healthy" they are downgrading higher priced quad cores to lower price tri-cores.

http://www.amd.com/us-en/Corporate/InvestorRelations/0,,51_306_15891,00.html?redir=IR0010

(The foundry company presentation, 3rd from bottom, has a lot of the fab/process 'info')

Tonus said...

"- 32nm plans are a bit vague - AMD has some crazy picture which shows 1st test chip Q1'10 and "qual" (whatever that means) at end Q2'10. Not sure if that means add some additional time for actual 32nm product (i.e H2'10)? On a separate product presentation, they show no desktop/notebook 32nm product until 2011."

And here I thought that they were getting ready to release 32nm chips in 2009. *smirk*

Tonus said...

AMD's Fusion, 32nm moved back to 2011!

AMD plans to release a quad-core mobile CPU in 2010 now, and will wait until 32nm for it's CPU/GPU hybrid and its Bulldozer CPU for the desktop.

Intel plans to release a CPU/GPU hybrid in late 2009. Assuming that they meet their target, this could be another blow to AMD, and in a market that is growing very very fast.

Anonymous said...

Not if it's an AIG style bailout where the gov't takes an 80% stake in the company. Shareholder's will be wiped out. Personally, I'm staying out of this market right now. It's just too volatile, cheap is relative.

Sheesh, don't remind me :). I'm cotrustee on a $500K annuity account with AIGA, a subsidiary of AIG, and back in Sept. I was ready to pull the plug and lose $10K in interest since the principle is only insured for the first $100K. However, now that the Feds are in for $150B, I feel pretty confident they won't let AIG go under and so will wait the last year.

I also agree with Sparks that the feds won't let GM or Ford go down the tubes. Haven't been following the stock price but it might be a good bet. Stocks I'm avoiding for now are Apple & Google and V (Visa) since I think they are more dependent on the "luxury" than basics purchase categories, esp. Internet advertising in Google's case. I'm sure they'll jump back up once the turnaround occurs (lots of pent-up spending to occur?) but I don't think they've bottomed out yet.

InTheKnow said...

I'm sure they'll jump back up once the turnaround occurs (lots of pent-up spending to occur?) but I don't think they've bottomed out yet.

If you buy the theory that this downturn is going to be similar to the one in 1873 (due to similar causes), then we are a long way from the bottom. That event took the US 4 years to recover from and Europe 6 years.

I fear we may be in for a long haul.

Though as I stated above, those with cash in hand were able to prosper.

Drawing on experience in my own lifetime, this reminds me a lot of 1976, when Jimmy Carter swept into the White House, also gifted with a democratic congress if memory serves correctly. During his campaign President Carter created a metric he called the "misery index". It was a combination of unemployment and inflation as I recall.

Four years later the misery index was even higher than when Carter was elected and he and his congress were washed out by Reagan and the republicans. If this downturn mirrors the historical precedent, President Obama may be a single term president.

Anonymous said...

AMD's Fusion, 32nm moved back to 2011!

Ok, may be over-reaching a bit here - but looking at the roadmap, I wonder if AMD is having issues doing an MCM approach.

The initial plan for fusion was Gen1 to be an MCM and gen2 to be monolithic. It's hard to tell, but I'm guessing that the 2011 GPU/CPU is monolithic.

I also no longer see an MCM 8 core chip (though the server roadmap is not extremely clear)

I also find it pretty funny how many times profitability and the goal of consistent profitability was talked about at analyst day. The fact that they have to state this so overtly and repeatably is indicative of how AMD has been run over recent years. The goal of consistent profitability should not have to be a goal - it should be an expectation. The more it is stated, the more it makes people wonder how much of a priority it was in the past. For a startup, things are different - you may choose growth or product development over profitability, but for a company as old as AMD, people should not have to be reminded. Unfortunately for AMD, they need to do this; because Ruiz went on his market share at all cost binge.

Anonymous said...

Tick Tock Tick Tock, is that Arab Micro Devices death I hear?

Lets see INTEL made how many billion last quarter. Now imagine if they took that billion divide by say 100 million CPUs that is a serious price cut. INTEL would still break even. AMD couldn't make money in good times WTF are they going to do if INTEL has to cut prices to move the invetory? Nothing is more expensive then an empty multi billion chip factory. I expect INTEL to drop the prices on CPUs to try and move inventory.

IN this day of 60 dollar oil even the sugar daddy is hurting.

I'll wager AMD won't see the light of day in 2010. 2009 will be all about spin control and squeezing. The only crying will be Arab's who will have whished they had invested in something else.

Its the worst of times for the sorry men in Green.

Tick Tock Tick Tock the clock has run out.

Where the F is Scientia these days, LOL. Did he die in his basement or something?

Tonus said...

Heh heh

On January 8th, 2009 AMD will launch the Phenom II X4 in two models... the 920 and the 940. :)

(No word on availability of the 965!)

Anonymous said...

Curious numbering scheme by AMD, or not. Perhaps they believe Phenom II will compete with equivalently numbered Nehalems?

Anonymous said...

Typical AMD gimmickery:

2.6GHz - 910
2.8GHz - 920
3.0GHz - ???

Well obviously it is 940! And what happened to 930, other than a marketing need to put more separation between top bin and the next bin? And where is 'the customer demand' for the 2.0 and 2.2GHz desktop chips now? AMD is fortunate that customer demand has shifted to higher clocked quad chips at the same time they are able to make them.

It's good to see AMD focusing on the quad desktop space as the desktop market is growing so rapidly and the quad marketshare in that area is so large! Though it looks like they may actually have a dual core mid'09 a mere 18 months after the original Dec'07 roadmap, or the revised Q1'08 forecast or the re-revised Q2'08 forecast...

It is also humorous to see AMD fans rejoicing over chips 2.8,3.0GHz that they were expecting in late 07 or the beginning of '08 on 65nm.

Anonymous said...

Nov 8 comment:

Well I was a little off on the timing, but here we are with Yahoo @$12.2/share (vs the MS offer of $33 back in late Jan) and Yang publicly BEGGING Microsoft to do a deal...

And Balmer says no, just as publicly ('It's in the past')! Perfect! The pressure on Yang will cause him to step down or make the first offer (perhaps thru Icahn) to MS... and MS will be bargaining from a position of strength. And Balmer or may not say "Do you like Apples? I said do you like Apples? Well how do you like them Apples"?


And in the news today, less than 1 week later Yang steps down! This clears the way for what most analysts believe will be Micro-Hoo. I wonder how big the parachute will be and whether this means he dodges the eventual shareholder lawsuits?

Rather sad - begging for a deal, getting laughed at and then being forced to scurry away into the darkness.

Almost like overpaying for a graphics company, begging for money from Wall St (Morgan Stanley), begging for money from the Middle East, and then having to spin off fabs to secure even more money from the Middle East...and calling the strategy "asset smart" - did I say that out loud?

InTheKnow said...

An interesting article in EE times on AMD's 45nm transistor performance.

In particular this paragraph at the bottom of the third page caught my eye.

The transistor drive current for AMD's 45-nm devices is much lower than that of the Intel HKMG transistors. But power consumption is quickly becoming a high priority for server chips. AMD's transistors exhibit very low channel leakage. Our transistor benchmarks indicates that leakage current is less than one-third of the value measured on AMD's 65-nm process. It's also significantly lower than the Intel 45-nm HKMG process. In fact the Ion/Ioff ratio for AMD's PFET is nearly 10 times better than that for the Intel PFET.

If I'm reading this correctly, then IBM's (let's be real about whose technology this is) transistors will continue to show better idle characteristics than Intel's. So we should expect to see AMD show better idle power performance than Intel unless Nehalem's power management can compensate for the difference.

Anonymous said...

It is also humorous to see AMD fans rejoicing over chips 2.8,3.0GHz that they were expecting in late 07 or the beginning of '08 on 65nm.

What's really humorous is how giddy the green fanbois get over another "simulated" performance marketing blurb on a yet unreleased CPU. One would think they learned a lesson about a year ago with Barcie, but apparently not. Also the fact that Intel releases the Ci7 and AMD releases a marketing blurb :)

Anonymous said...

Where the F is Scientia these days, LOL. Did he die in his basement or something?

Actually he has crawled out of his basement and is now posting BS on UAEZone once again:

AMD needs about 30% more capacity to make chipsets for every cpu. AMD might need 10% to make GPU's. This means that AMD's projected capacity will exceed the expected rate of growth of the market. And, this means that once again AMD plans to take share from Intel.

The above after humiliating himself so profoundly with his "expertise" on Asset Light :)

Abinstein posted a reply where he thought the Foundry would take in outside fab orders, and Sci immediately got defensive, wanting to see links, etc. and using his "do you honestly believe" innuendo that casts doubt on the other person's honesty and intelligence.

Sci should just stick to what he knows and leave the fab and business analysis to his betters (which would be most posters here :).

Anonymous said...

Sci should just stick to what he knows and leave the fab and business analysis to his betters (which would be most posters here :).

Funny, didn't see any links bacing up his 30% and 10% claims! His 2010 Wafer start claims are also misleading as he obviously used the INQ which used the analyst day foil which simply did capacity by year (which is the end of the year). The average over the course of 2010 will be somewhere between the end 2009 level and the end 2010 level.

Actually AMD (via AMD foundry) has stated that 32nm is where they will look for outside orders and Meyer, when asked at the earnings call if F36 would be SOI and F38 would be Si, said that is a bit oversimplified but a close way of looking at things.

The problem is people see F38 as excess CPU capacity - I think it will be largely foundry capacity. I'm sure AMD will have some SOI swing capacity via F38, but once again people are reading into the plans for what they want to believe. The NY fab is not producing anything until 2012 (from AMD analyst day foils on capacity), so are we to believe the foundry will not be doing any outsourcing until 2012/2013? If that were true the foundry company would not survive (or would be even less healthy than ttoday.

It is also humorous that he keeps making the mistake that a shrink effectively doubles UNIT capacity - people continue to make this mistake even though 45nm parts are only (I think)15-20% smaller due to extra cache. 32nm will be no different - there will be more cache, more transistors for other features which will keep die scaling below 50% (as one 'theoretically' can calculate)

Tonus said...

re: market share-- how plausible does that seem? Let's say that this statement is true:

"AMD's projected capacity will exceed the expected rate of growth of the market."

Does this also apply to Intel, or is Intel likely to be capacity constrained in the coming year? I know that they were unable to meet the (unexpected) demand for Atom. Would that affect availability of their remaining CPU lines?

If Intel can't make enough CPUs to meet demand, this is good for AMD, as it can help with both sales and ASPs, right? Otherwise, just having the option to make more CPUs will not, in itself, mean you gain more market share. I think his next sentence is curious:

"And, this means that once again AMD plans to take share from Intel."

Does this mean that he feels that AMD will once again go after market share (possibly hurting ASPs to do it)? Or is he attempting to imply that AMD *will* take share, while wording it in a fashion that allows him an out if AMD isn't able to do that? (ie, "I didn't say that they would take share from Intel, just that they planned to.")

I would assume that AMD always plans to take market share from Intel, or at least they must have that as a goal.

Tonus said...

nonny: "The above after humiliating himself so profoundly with his "expertise" on Asset Light :)"

Apparently he was also under the impression that Intel was going to make socket 775-compatible Nehalem CPUs, which would reduce memory performance. I agree that moving Nehalem to socket 775 would hurt memory performance. :)

Anonymous said...

Tonus: Apparently he was also under the impression that Intel was going to make socket 775-compatible Nehalem CPUs, which would reduce memory performance.

I just put that down to his temporary "vacation" from posting or admin-ing at UAEZone, which coincidentally happened exactly when AMD revealed his "asset light" analysis was 100% incorrect :). Everybody is entitled to a 'brain fart' once in a while :)

However his "AMD plans to take marketshare from Intel" is fairly laughable. I just don't see AMD's current and near-future product line being that competitive. However since it takes such a long lead time to get a fab up and running, maybe AMD is thinking Bulldozer might turn their fortunes around. Of course, AMD/TheFoundry is banking on the current recession being short-lived; otherwise The Foundry will have lots of expensive equipment and employees mothballed.

Funny how Intel has drastically revised their Q4 estimates downward, yet AMD is playing a waiting game, I guess to keep their stock from completely tanking. I guess you could say AMD is banking on not tanking :).

Tonus said...

Well, if they price their CPUs to sell, or take advantage of shortages if Intel can't produce enough CPUs, then they can make up some market share without competing at the high end. That's really what I am wondering. If we assume that AMD will have sufficient capacity to meet and even exceed overall market growth, what would the prospects be for that happening?

Does anyone feel as if Intel will have trouble meeting market demand (or is that even something that we can know with any degree of reliability)? Would cutting ASPs further be useful if it means more market share, or would it be suicide? Their prior attempt at grabbing up market share is what led them to their current crisis, isn't it (in part, at least)?

I'm curious as to what the view is among the regulars here. I get the impression that AMD and the Foundry Co are probably not going to be sitting on any excess capacity. I guess it's time to see if all of my time reading here has made any dent in my thick skull. :)

Anonymous said...

Well, if they price their CPUs to sell, or take advantage of shortages if Intel can't produce enough CPUs, then they can make up some market share without competing at the high end. That's really what I am wondering.

This would be true if the market was growing or if Intel suddenly lost capacity. Intel is predicting a shrink in revenues (which presumably would be flat to down unit sales unless ASP's fall off a cliff); so if AMD is not competing at the high end then this means there is just more competition at the low end.

This can result in a few things:
1) Buildup in inventories (which neither company will want to do)
2) Increased price cuts (in my view, likely); this may also spur growth a bit
3) One or both companies cutting production capacity temporarily (while this may happen, I doubt it will be a significant amount)

The other thing lost in the discussion is the time period - Scientia is talking about capacity coming on line starting (to a small degree according to AMD analyst day) in 2009, with the majority of the output ramping in 2010. Who know what the economy will be like then. This is also likely going to be BULK Si capacity (for chipset, GPU's and the beginnings of foundry capacity) - so I don't see how this significantly feeds CPU volume. Presumably there is also increased mix toward quad core, so while the shift to 45nm may help some, it will be blunted by an increasing mix away from dual core Athlons (i.e you can't simply use wafer starts to project capacity)

There are just too many holes, assumptions and a complete lack of understanding in his propositions (he also essentially treats SOI and bare Si as fungible capacity)

My sense is he waited long enough to hope that the foolishness of his Asset Smart "knowledge" is forgotten and he can once again pawn himself off as an expert in the area. He has fallen back in "extrapolation mode" and doesn't have the foundational background to understand the foolishness of some of his analysis.

Anonymous said...

Does anyone feel as if Intel will have trouble meeting market demand (or is that even something that we can know with any degree of reliability)?

I doubt it; demand is soft right now and with the exception of possibly some short term mix issues (like Atom), Intel also has additional capacity as they finish the migration to 45nm.

Would cutting ASPs further be useful if it means more market share, or would it be suicide

I don't see Intel gaining much market share as with the PC market flat to shrinking the only way they do this is by AMD shrinking. (If the PC market was growing, intel could effectively gain share by "getting" most of the new growth). At this point AMD's F36 is underutilized so unless they take some capacity offline, there will be excess supply.

So any Intel price cuts will just punish AMD, as AMD will be forced to respond. I would hope AMD is smart enough this time around not to start this game up again and will only cut if they need to respond to Intel - that is if we are to believe that "sustained profitability is our top priority". If you see AMD starting the cuts then it's "meet the new boss, same as the old boss" with the market share at all cost mentality.

Also keep in mind in the future AMD is also paying the foundry company margin on top of wafer production cost, so it makes AMD even worse off in a price war.

Given Intel's revenue warning, unless they start building huge inventories, I don't see Intel starting a price war - even though it would have much more of an impact on AMD then Intel.

InTheKnow said...

he also essentially treats SOI and bare Si as fungible capacity

There you go discounting AMD's APM advantage again. Will you never learn. :)

Anonymous said...

Gotta love Fudzilla! They're saying maybe a 32nm shrink for AMD before Bulldozer in Q4'09.

Of course AMD's own slide from analyst day has first test chip on 32nm SOI in Jan '10... so it'll be interesting to see production chips come out PRIOR to the first test chip!

The guy make Scientia's analysis look good!

Tonus said...

Q1 2009 (Nehalem) Xeon pricing

Quote from a post at the Aces Hardware forum: "[the 2.53GHz/80W: $744 Xeon] Will be squaring off against $698 2.7Ghz/75W ACP Shanghai. I suspect this won't look so good for Shanghai."

I am under the impression that AMD and Intel measure power consumption differently, so I'm not sure how the 75W/80W figures there compute exactly. But Nehalem has been scoring well in Spec and if they're able to address issues of power consumption and heat, then it is indeed a potentially big problem for AMD.

Anonymous said...

I doubt it; demand is soft right now and with the exception of possibly some short term mix issues (like Atom), Intel also has additional capacity as they finish the migration to 45nm.

I'm starting to believe various reports that this 'recession' is fundamentally different from previous downturns (with the exception as point out here first, of the 1873 crisis). We should get a good idea of the market conditions after this quarter I guess; however the experts at Intel have already made their predictions. I sorta think AMD may have been blowing smoke up the UAE burquas in order to get them to buy out the fab part since AMD could no longer afford it anyway.

Anonymous said...

AMD's stock has been flirting with the sub-$2.00 range all day long.. Guess we'll find out in an hour if it achieves its lowest close since October 1990 ($1.94).

I was just thinking that if UAE ever moves the new foundries to the Emirates, then Intel could hire the Somali pirates to really put a dent in AMD's business model. Just imagine AMD having to pay $2M ransom for a shipload of CPUs probably only worth $1M :).

Tonus said...

It appears to have closed at $1.91.

Anonymous said...

I guess it's time to choose whether to throw money at it or not. Somehow I don't feel too lucky about it, so I won't make Hector's day :)

SPARKS said...

Look out below! The AMD Piano is falling!

http://moneycentral.msn.com/detail/stock_quote?symbol=AMD

SPARKS

Anonymous said...

Geez - a dollar sixty-three as I type this at 3:10PM EST... That's less than what a gallon of gas costs here in the Washington DC area ($1.65 this morning).

Well at least Deneb samples can clock up to 6GHz on LN2, so the fanbois are happy and claiming Ci7 is about to get B-slapped with a wet fish...

Tonus said...

I was glad to see the early OC results for Deneb. 4GHz air, 5 and 6GHz with more extreme cooling. It could make for some very interesting choices for the hobbyist looking for a relatively cheap platform to overclock.

Although I have to wonder... what would the reaction have been if Intel had shown early samples of a new CPU running at 4/5/6GHz and not making CPU-Z screenshots available, or allowing any other verification methods?

Anonymous said...

The overclocks were impressive, but I'm not sure it is practical - while many quote the temps at 4GHZ, they seem to not mention it took 1.6V to break 4GHZ, and I can only manage the voltage applied to get to 6GHz on LN2 (1.9V+?), not to mention that all this could do is boot - it was not stable beyond this. Where are the cries for Prime 95? :)

All in all, good news - but I don't see people operating at 1.6V for very long.

It's funny that people seem confused or projecting that AMD should be able to release much faster stock chips based on this. When you look at the voltages (and not temps!), it is not surprising at all.

Anonymous said...

http://www.xtremesystems.org/forums/showthread.php?t=205779&page=12

4.3GHz Core i7 on stock cooler at 1.344Volts

http://www.xtremesystems.org/forums/showthread.php?t=205779&page=11

4.2GHz Core i7 on air cooler(non stock) at 1.4Volts

So a little perspective on Deneb - hitting 4GHz @1.5-1.6V is good but not earth shattering. Hitting 4GHz on air is good, but again not earth shattering. It will be nice to see some actual screenshots, it's surprising (must be the cynic in me) that with results "this good", there is minimal information. From my perspective, I'd like to know what the max OC is under 1.5V (preferably around 1.4V). Anything above that(and above 1.4V for me) is just for show - which while fun is not really that useful.

pointer said...

http://www.xtremesystems.org/forums/showthread.php?t=205779&page=12

4.3GHz Core i7 on stock cooler at 1.344Volts

http://www.xtremesystems.org/forums/showthread.php?t=205779&page=11

4.2GHz Core i7 on air cooler(non stock) at 1.4Volts

So a little perspective on Deneb - hitting 4GHz @1.5-1.6V is good but not earth shattering.
...



you just can't compare voltage just like that because AMD and Intel use different process .. thus we need to know what are the safe working voltage range for these 2 processes then only we can do some meaningful comparison ... having said that .. i didn't bother to look it up ... :)

Anonymous said...

"thus we need to know what are the safe working voltage range for these 2 processes then only we can do some meaningful comparison ..."

If your're worried about electromigration as the main long term reliability issue, then you can - there may be some subtle differences (shunting layer?), but Cu is susceptible. If you are worried about oxide reliability then yes, you have a point.

The safe working range is also somewhat subjective unless both companies are using the same criteria for what is an acceptable failure rate. Do they both use the same temp ranges over that voltage range, do they use the same chip lifetime criteria, etc...

I wouldn't really care if AMD claims the safe range covered some of the voltages in question - operating a chip for an extended period of time at 1.5-1.6V does not seem like a good thing to do.

My point is people are getting jazzed up about 4GHz on air, but there is limited info on the Vcore to get there - it seems to be in the 1.5v range on the AMD chip. On the Intel chips people are going OVER 4GHz at 1.3-1.4V, I even saw one on xtreme systems claiming 4GHz in the 1.25V (or so) range (rhough I'm not sure they had a CPU-z shot to back it up). Some folks are also getting the Core i7 920 chip (the lower bin) to 4GHz and up as well

The 6 GHz data point is totally bogus and sounds nice for marketing, but first we are talking LN2 and second the voltages appear to be in excess of 1.8V. It also states they were only able to boot the OS.

The water cooling OC would be an interesting data point, but again the Vcore to get in te 4-5GHz range appears high to me (1.6V or so).

A word of caution - not so long ago AMD was doing a K10 demo at 3GHz on air cooling about 3-6 months before the Barcy launch.

Anonymous said...

A word of caution - not so long ago AMD was doing a K10 demo at 3GHz on air cooling about 3-6 months before the Barcy launch.

Yeah, that turned out pretty lame :). About on a par with the infamous "dancing in the aisles" Fuddo post.

I suspect Deneb is going to be a big improvement for AMD on the desktop, but it still won't take the high-end from Intel. I really want to see the side-by-side comparision in a multi-GPU setup with Ci7, as I'm thinking of a new build around February or March, as I get my tax refund around that time. After all, I gotta keep up with Sparks :)

SPARKS said...

“My point is people are getting jazzed up about 4GHz on air,”

“A word of caution - not so long ago AMD was doing a K10 demo at 3GHz on air cooling about 3-6 months before the Barcy launch.”

True,----very true, on both points.

I’ve been running the QX9770 (as I am now) @ 4 gig @ 1.425V since April. So what’s the BFD?

As you implied, AMD’s pump the neighbors’ cat marketing tactics may work in the short term. However, much remains to be seen with consumer grade product, with daily day to day stability, under real world operating dynamics.

Sure, QX9770 will go to 4.27 with a bit more V. But, there is a sharp edge at this precipice. I don’t like it there; neither does the chip nor does XP. This is not to mention the horrors of:

“If your're worried about electromigration as the main long term reliability issue, then you can - there may be some subtle differences (shunting layer?), but Cu is susceptible. If you are worried about oxide reliability then yes, you have a point.”

(Every time he talks like that I’m ready to hit the BIOS and kick the multiplier back a notch)


All said Pheromones 2 needs to get through QX9770 BEFORE it could even look to compete with i7, let alone an i7 965EE gorilla. Ironically’ wasn’t AMD the one who proved that core speed wasn’t everything during the P4 955EE era a few years back? Why the about face? Further, what I’d like to know, where are the benchies to go with all this marketing fluff?


They dare not, not now, and it all horseshit for die hard fanbois.

SPARKS

InTheKnow said...

what I’d like to know, where are the benchies to go with all this marketing fluff?

As would I. In particular, I'd like to see some spec scores. Since Nehalem should really shine in the server space spec would seem to be the appropriate benchmark.

If Shanghai is that good you would think the OEMs would want to get the spec scores up as soon as they can. Maybe they are working on it and it just takes a while, but the longer it takes to see something, the less competitive Shanghai starts to seem.

But if you dig enough, you can find some bits and pieces, like this one. It is a sneak peek at Nehalem XP (at 2.8 GHz) and it should rule the server space.

Compared to the current Xeon you get:
That is because, in the SPEC CPU2006 benchmark, Intel's pair of Nehalem-based processors managed to achieve a score of 160 points, which is almost double of what the Penryn-based Xeon dual-socket platform can achieve, at 90.

And compared to Shanghai you get:
Moreover, this score places Intel's upcoming Xeon processors above a pair of AMD's new 45nm-based Shanghai chips, clocked at 2.7GHz, which can only get 105 points. Also, the SPEC CPU2006 result reveals just how close, in terms of performance, two of Intel's Xeon processors are to four of AMD's Shanghai Opterons, which can score 190 points.

Tonus said...

I hadn't really considered the voltage, I guess it's just been a while since I bothered with overclocking. But yes, it seems as if the i920 overclocks very well without large increases in voltage. Motherboards.org got theirs to 4GHz with an increase to 1.344v, and HardOCP was able to overclock theirs to 3.5GHz without increasing the voltage at all. Their CPU would not handle 4GHz stable, and required 1.375v to run at 3.8GHz.

So it seems as if they got a chip with relatively poor headroom, and that still provided a "free" 833MHz overclock. Time will tell how well Deneb OCs, I know that many sites will give it a shot.

SPARKS said...

“In particular, I'd like to see some spec scores. Since Nehalem should really shine in the server space spec would seem to be the appropriate benchmark.”

Exactly! While the Penryn series has a lock on desktops, now and into the next generation, I’d like to see them taken this new bad boy out for a spin into the space it was designed for, 4P and 8P.

As I’m familiar with desktop specs and performance, I would also like to see how i7 will measures up in the server and HPC space (encryption and scaling) which many of you are much more knowledgeable.

Tonus, Vcore is very critical to successful overclocking, as you know. But a well designed motherboard (as ITK will tell us) is a key factor, too. Forget the traces and how they interact with each other; the newer 8-Phase Vcore power regulators that practically surround the CPU these days are paramount to stability.

If you look at this link you’ll see the little black boxes (inductors that smooth ripple). Immediately behind them and underneath the heat sinks are the regulators. These suckers get HOT! With interconnecting snaked heat pipes and a multitude of heat sinks that sport miniature clip on fans, this is all mandatory to keep Vcore stabile to one thousandth of a volt with a minimum of fluctuating ripple! Amazing!

They bitch about 300~400 dollar motherboards, and then cry when they can’t get the same stable over clocks on a 150~200 bread and butter board. I have had great success with this board; in fact, it’s the best I’ve ever owned.

http://www.hardwaresecrets.com/article/490


Actually, I’m taking a hard look at this one presently for a new build up come April.

http://www.anandtech.com/mb/showdoc.aspx?i=3428

It seems buying this stuff is the only way you can cut through the biased bullshit these days. In fact, there was no mention (or very little) of QX9770’s performance until Nehalem samples were being released very recently. The bastards had something to compare its kickass performance to.

All said, like anything else, there ain’t no free lunch, but I suspect you already knew that.

SPARKS

SPARKS said...

ITK-This Neoseeker article/review speaks volumes. Let me draw your attention to the conclusion of the article that speculates Core i7 Xeon’s will do exceptional well. I’m not familiar with sever benchmarks; however, this guy seems to have a handle on them. He sounds very impressed.

http://www.neoseeker.com/Articles/Hardware/Reviews/nehalem_core_i7_review/31.html

Do me as favor, when server benchmarks do surface, please point us in that direction and give us (me) some analysis. I’d love to see some direct comparisons to Core2 results and Opteron results.

SPARKS

Tonus said...

Ouch! I don't know how I missed this one. HardOCP's take on Deneb vs i7 overclocking...

"But honestly, after getting all riled up about Phenom last time and getting left with my wiener in my hand, I will wait in giving any kind of opinion on the new Phenom II until I know for sure. The fact is that I just bought two Intel Core i7 920 processors less than $250 each after tax, title, license, and destination charge. Core i7 processors overclock obscenely easily and will eat anything AMD currently makes for lunch, and possibly dinner too. Did you notice we just left them out of our Core i7 articles? There was just no sense even spending the resources doing the benchmarks."

Maybe he's just a Paid Intel Pumper (hereafter known as PIPs), just like every other site out there!

SPARKS said...

“Maybe he's just a Paid Intel Pumper”


I don’t think so. In fact, a few years back he was one of guys (sites), like so many others at the time, who swallowed everything Arab Micro Devices handed out on a plate. In fact, that site, among others, was the reason I came here. (Lucky for you guys, eh?) I couldn’t get accurate product info. Most sites at the time where cheering on the “Scrappy Little Company’s” short lived success even AFTER the monumental release of C2D back in 2006. I’m sure many of you recall.

In any case, most sites lost serious credibility, especially that idiot who works for the INQ, Charlie D. “Dancing In The Isles”, cheese. I believe it was Digitimes who was the first to say enough was enough. Other’s then followed suit one by one. The straw that broke the camels back was AMD’s unconditional NDA that most sires refused to sign concerning Pheromones performance numbers. They just walked away pissed off.

Currently, an old cliché can describe most sites other than that other idiot Fuddie.

‘Once bitten, twice shy.’

HardOPC, like most sites, are taking a wait and see attitude and you can’t blame them.



SPARKS

Tonus said...

I was joking, sparks. :) I was just amused at seeing so many AMD fanatics summarily dismiss site after site as biased or bought off.

I'll bet that Kool-Aid stock is doing just fine these days!

SPARKS said...

“I was joking, sparks. :)……….”

Sorry about that. I used to get so annoyed with the so called “tech sites” for their unbridled biases, ‘till I found this place. More importantly, the education one could get around here, plus the interaction with some of the brightest people I’ve ever known, makes this place a refuge for someone like me with half a brain with a modicum of objectivity.

And………..I get to keep the QX9770!!!!

OXOX


SPARKS

SPARKS said...

Hey! Where the hell is everybody? Happy Thanksgiving, all, you clever SOB's!

SPARKS

Anonymous said...

Well Scientia has stopped writing misinformation on his blog... but that doesn't mean he hasn't stop attempting (unknowingly?) misinforming folks on UAEzone...

His latest "nugget":
Intel in contrast, is just getting its feet wet with quad manufacturing, has lost both the die size and DFM advantage, and is experiencing increased manufacturing costs due to having to use double masking. Intel's double masking pretty much neutralizes AMD's greater layers and SOI wafer cost. On top of that it very much appears that the 750 SB and newer 45nm process have neutralized nearly all of Intel's previous power draw and clock advantages.

It's amazing to see how in just 3 sentences, he can lock in so much misinformation. Where to start?
1) Immersion litho vs double masking - immersion tools are ~2X the cost and operate at a slower throughput. If the throughputs were the same, buying twice as much at 1/2 the cost is nearly a wash financially. It is a bit worse due to fab costs, chems, etc but then you have to factor in immersion litho is SLOWER than dry litho so that hurts it from a cost perspective. I have seen arguments in both directions as to which one is truly cheaper. My personal view is that this is not a cost decision but a manufacturability/risk one - you are trading off the risk of a double patterning process with the risk of an immature tehcnology (immersion). Given Intel is ahead of the curve on the tech node transitions it seems pretty obvious you would choose the mature tool.

2) SOI substrate cost is SIGNIFICANT. In the early days it was ~15-20% cost adder to finished wafer cost; I don't have any recent pricing on SOI wafers, but I'd guess it would be at least 5-10% cost adder (probably closer to 10%).

3) An extra metal layer is significant: This is an ILD deposition, 2 litho steps (though a lower tech, cheaper litho), 2 etch steps, 1 etch stop step, 1 barrier seed step, 1 plating step, 1 CMP step, and numerous clean steps. While these can be shared with other tools used on other metal layers - the extra layer will trigger some new tool purchases (higher cost). You will also consume more variable costs (chems/gases, labor, etc). Then there is also the factor that this is another layer for yield hits (probably not huge, but not irrelevant)

3) Quad manufacturing? What the heck does that mean? Intel is simply producing bigger die - something they have done in the past with some of the really large server die. He makes it sound like going to quad core means you alter your manufacturing process!?!?

4) Lost the DFM advantage?!?! :) Remember the days where he dismissed DFM as garbage/hype? Then when AMD started PR'ing this, he said "oh it will be significant and this will reduce the advantage" - this is the magic bullet on why Intel was better! He makes it sound like DFM is a one shot deal and you simply choose to implement it or not implement it. DFM (design for manufacturability) is an EVOLUTION over time - the more you do it, the more refining you can/will do. AMD suddenly deciding to pay attention to it doesn't mean they will catch up... nor does Intel using it mean they will stop continuously refining it over new products and tech nodes. I don't even think he has a clue what DFM really is/how it is implemented - but he feels more than confident editorializing about it!

Same old, same old - only now he posts in the safe confines of UAEzone, because he knows what will happen, cough, Asset Smart, cough, if he tries to blog in 'public' again.

Anonymous said...

Immersion litho vs double masking - immersion tools are ~2X the cost and operate at a slower throughput.

I'll just comment on this.

If we're comparing Penryn to Nehalem....

1. The (patterning) process is identical. Any disadvantage due to double patterning isn't new.

2. Double patterning *is* limited to one layer (poly). AMD needed to go to immersion for all of their criticals. You be the judge as to who best has DFM on 45nm.

3. Intel grossly underestimated their ability to manufacture these chips. That they can exceed commits on products delivered with a delayed fab startup is a testament to that. (and the skill sets of the engineers at PTD/D1D.)

InTheKnow said...

I think we are letting Scientia lead us down the rosy red path here and are missing one of Intel's biggest advantages. That is the virtual factory system. The system itself may have issues but it brings something to the table that AMD can't possibly match.

SCALE

So while AMD has their fab running around 30K wafer starts a month, Intel will have 3 fabs running around a combined 75K wafer starts a month.

This gives Intel a learning edge that AMD won't have. And no secret APM/DFM sauce can offset that.

With volume you see subtle defect modes show up more often. These defect modes tend to be hard to pin down and need to be seen several times to find and fix them. And you have the engineering resources of 3 factories working on solving these problems. Not just the resources from one factory.

But I doubt you will see that analysis from Scientia, because he has never been closer to a fab than the photo's in FABTECH.

Anonymous said...

So while AMD has their fab running around 30K wafer starts a month

I think you are quoting the eventual potential output of F36, right now I think it is at ~25K WSPM (it was in the analyst day slides), AMD shows it hitting final output levels in 2010. Don't mean to be picky - while 5K is a small #, for AMD it would be an ~20% output increase!

With volume you see subtle defect modes show up more often

This is very true...I think the fact that 45nm is more or less 65nm with immersion litho is why 45nm for AMD appears to be starting out relatively well (assuming the projected clocks and timelines are true) - the learning curve isn't there as much as other nodes. The real interesting thing will be the 32nm startup as that will be high K introduction for AMD and that is where scale/volume is critical in terms of sorting out the new yield/process/reliability issues associated with High K gates oxides.

SPARKS said...

“On top of that it very much appears that the 750 SB and newer 45nm process have neutralized nearly all of Intel's previous power draw and clock advantages”

This got me thinking and searching.

http://processorfinder.intel.com/details.aspx?sSpec=SLBCJ

According to INTC i7 965 is rated at 130W. This doesn’t mention the core (s) literally shut down when idling.


Even when compared to QX9770, they’ve managed to reduce advertised TDP by 6W at the same speed. But, QX9770’s TDP in only 103W @ stock speeds.

http://www.tomshardware.com/reviews/intel-core-2-extreme-qx9770,1736.html


Pumping up the volume @ 4 gig, however you will find yourself 170W territory.

As for clock advantages, I say bring ‘em on! Enough paper launch power pimping. Let’s get some REAL hardware, a solid launch, and let’s go for it! Forget Core i7, I say I have lived with a year old QX9770 that will kick Pheromones II ass back to the Middle East, clock for clock, watt for watt, at any speed!

Sooooo…………..The quest begs. WTF is he talking about????


SPARKS

Tonus said...

I think it was guru or nonny who explained once that scientia's analysis typically assumes very favorable developments for AMD and very unfavorable developments for Intel, which is why his predictions were more accurate when AMD was doing well, and less so since Intel reasserted itself with Core2.

So I guess his recent comments just follow that trend, where AMD catches up to Intel in process tech, and where awesome new ideas like DFM help them become more competitive and possibly put Intel back on its heels. This fits with the general tone of the comments there. Has anyone noticed that when AMD adopts a technology that Intel has already implemented (HKMG, for example) it's a sign that they're catching up... but when Intel adopts a technology that AMD has already implemented (native quad, IMC, HT) it is a sign that AMD is catching up, since the learning curve for Intel will slow it down?

While I enjoy the discussion of the companies and the industry and the technology, I tend to refrain from making predictions, because I understand that I just don't know enough about what is going on to make an intelligent analysis. I prefer to let others provide the analysis and predictions, and then let time and circumstance show who was right. Thus, I never have to worry about a bruised ego. :)

SPARKS said...

GURU, from this Super GURU----

“Intel is working on design techniques that can be coupled with different lithography solutions, he said, noting that proprietary forms of phase-shift masks (PSMs), design for manufacturing (DFM) and optical proximity correction (OPC) developed by the lithography team have been able to extend immersion lithography to the 22 nm node.”

Pixilated patterning, nice.

Do you think Dementia knows ANYTHING about this?

http://www.semiconductor.net/article/CA6553758.html

"Now, Bohr said it is “fairly routine” to be creating 32 nm test chips at the company’s development fab in Hillsboro, and said the next public milestone will be to demonstrate working 32 nm microprocessors. One advantage, he said, is that Intel can extend its high-k/metal gate technology to the 32 nm node and beyond."

Sure, AMD has evened up with INTC. Yeah, right.

(As a side note, he is right there with you on the EUV front.)


SPARKS

Anonymous said...

because I understand that I just don't know enough about what is going on to make an intelligent analysis.

Well you need to head on over to UAEzone, where lack of knowledge is generally preferred when making predictions, otherwise the predictions would not be so positive for AMD. I understand that it is a forum for AMD fans, who tend to root for the company to do well - but I don't understand why it means everything regarding Intel must be spun negatively.

You have Scientia also talking about AMD mobile being as popular as ever, yet neglecting to mention that they actually LOST market share in this segment and of the 3 main CPU segments, mobile is their lowest share with respect to Intel. With popularity like this who need enemies/competitors?

I'm still trying to figure out how Intel is playing catchup on "quad manufacturing". If anyone could explain the logic on that one, I would be forever grateful!

SPARKS said...

“ I'm still trying to figure out how Intel is playing catchup on "quad manufacturing".”


I think he’s referring to ‘native’ quad core manufacturing, which taken literally could be interpreted as true.


SPARKS

Anonymous said...

As a side note, he is right there with you on the EUV front.

Well Mr.Bohr may have just a wee bit more knowledge on these things, so in the event he does state something differently in the future, I would tend to trust him more than me!

I wonder what IBM has to say about EUV - after talking about some breakthroughs and actual patterning (I think on a 2 layer structure?) and an implication that maybe it would be 22nm - not much has been heard from them. Smells a lot like yet another 'researchy' announcement where it goes nowhere near production in the timeline that is hinted at (see SiLK, high K - 45nm, soon to be vacuum air gap, soon to be EUV on 22nm)

What I do wonder is if folks will be able to get 32nm on immersion only (without double patterning)...22nm would seem to have to go double patterning regardless.

That said I do wonder how extendable the integration scheme of Intel's highK/MG solution is beyond 32nm. Bohr may be taking liberties with the meaning of extendable. The films, techniques and materials may be extendable, but I wonder if the replacement flow (where you are filling ultra small features with the gate metal) is going to continue to work without voiding or other issues. At 22nm, you are talking about filling features that are probably ~18-20nm wide (or somewhere on the order of 50-60 atoms or so wide). With trigate looming, Intel will likely have to switch to a gate first flow (similar to what IBM will be doing on their high K process) - if not 22nm, this would seem likely at 15nm?

I still see a combined change of SOI, trigate and likely at the same time a switch to a gate first flow - this may be 22nm or 15nm. (I think in the past I have predicted 22nm)

InTheKnow said...

I still see a combined change of SOI, trigate and likely at the same time a switch to a gate first flow - this may be 22nm or 15nm. (I think in the past I have predicted 22nm)

My best guess for tri-gate was the 32-22nm time frame. So if they don't move to tri-gate by the next node, I missed. Given the structure of the trigate transistors, it is not at all clear to me whether gate first, or gate last would be preferential.

As to SOI, I seem to recall seeing something about Intel doing some sort of localized, in-process insulator technology. I'll have to dig around and see if I can find the link again.

SPARKS said...

Yup, ITK, here's some.

This article on ‘gate first” may be of some interest.

http://www.intel.com/technology/itj/2008/v12i2/1-transistors/3-processflow.htm

SOI, however, is something INTC never mentions, except “FDSOI is under investigation”.

The tri gate deal is gonna happen; they’ve got nice pretty pictures like these:

http://www.intel.com/technology/silicon/integrated_cmos.htm

Especially this nice little presentation, they even mention FDSOI.

http://download.intel.com/technology/silicon/tri-gate_foils_VLSI_0606.pdf


Ya sure gotta hand it to AMD’s R+D.

SPARKS

SPARKS said...

In one of my above posts I linked the TDP of QX9770 to INTC which rates the chip @ 103W @ a stock 3.2 GHz. The chip is rated for 130W.

Austin Computers has the Pheromone II listed here.

http://img377.imageshack.us/img377/5098/83783891tc1.gif

125W @ 3GHz selling for 500+ dollars, as a opposed to the Pheromones II 920 @ the same TDP selling for 471 dollars @ 2.8 GHz, go figure.

(G,- how are they binning these things?)

Give me a Core i7 920 for $300, plus a nice bread and butter X58 board,

http://www.newegg.com/Product/Product.aspx?Item=N82E16819115202

http://www.newegg.com/Product/Product.aspx?Item=N82E16813130216

I’ll get this setup, eat the memory,

http://www.newegg.com/Product/Product.aspx?Item=N82E16820145221


and blow Pheromones II out of the water. The difference in price is negligible ~ $100

It’s a no brainer, and another AMD miscarriage launch.


SPARKS

Anonymous said...

Sparks, you did realise that Austin Computers is an Australian etailer, and those prices are in Australian dollars?

See here for their Core i7 prices:

http://www.austin.net.au/ProductList/tabid/103/Search/1/ManuID/56/Default.aspx?Keyword=core+i7

i7-920: $517
i7-940: $997
i7-965: $1798

SPARKS said...

Whoops, there goes that theory! I'm shot. Hmmm, looks like AMD might have a bargaining chip.

SPARKS

Anonymous said...

Given the structure of the trigate transistors, it is not at all clear to me whether gate first, or gate last would be preferential.

The main issue I see is with the metal fill in the replacement flow. With trigate you are no longer filling a simple trench structure which will likely make it harder to fill (it may also make it harder to put down the workfunction metal, unless they are using ALD). The structure would essentially have a fin at the bottom of it - looking at it sideways you might think of it as a W, with the points flattened off, or essentially a... |_n_| instead of a simple.... |___| ...pictorially. The other issue is extendability of filling as a viable solution as the gates get smaller and smaller (regardless of structure).

The reason not to think of trigate for 32nm is that Intel rarely does a major change for one generation and trigate on 32nm would have would likely require a major re-work of the HK/MG vs simply shrinking it. Intel can probably get the performance gains from regular gate oxide scaling of the high K or modification to the high K material (I think they have stated that the solution is extendable to 32nm) and then use the performance gains of tri-gate when the highK soultion can no longer be scaled further.

The reason I throw SOI into the mix is that SOI will potentially enable some higher density alternative memory cells over the current 6transistor (or now 8 on Nehalem) SRAM cells. This would enable significantly more cache or smaller dies sizes.

Anonymous said...

Same old, same old - only now he posts in the safe confines of UAEzone, because he knows what will happen, cough, Asset Smart, cough, if he tries to blog in 'public' again.

Well it _used_ to be a lot safer for him :). But starting with enumae pointing out how some of his absurd comments conflicted with his blog postings, Sci's return from the grave may have been premature. I get the impression his credibility even on UAEZone is closely tracking AMD's stock performance - i.e., hovering near worthless :).

Anonymous said...

Tonus said...
I think it was guru or nonny who explained once that scientia's analysis typically assumes very favorable developments for AMD and very unfavorable developments for Intel, which is why his predictions were more accurate when AMD was doing well, and less so since Intel reasserted itself with Core2.

Sounds too astute an observation for me to have made :) so probably Guru said it. I just like pointing out all his mispredictions and leave it at that.

Has anyone noticed that when AMD adopts a technology that Intel has already implemented (HKMG, for example) it's a sign that they're catching up... but when Intel adopts a technology that AMD has already implemented (native quad, IMC, HT) it is a sign that AMD is catching up, since the learning curve for Intel will slow it down?

With all that "catching up" going on, why hasn't AMD actually caught up yet?? :) I think that we'll get a good idea of how Shanghai/Deneb and AMD's 45nm process are doing when the Q1 reports come out next April. If we see low volume and a lot of cripple-cores being sold, then I would think the rumor of AMD having low yields on 45nm might be true.

Anonymous said...

then I would think the rumor of AMD having low yields on 45nm might be true.

I would actually be surprised if they had major yield problems on 45nm - there is just not that much different from 65nm. On the other hand there were significant changes from 90nm to 65nm which I believe lead to thermal/clock issues (and yield issues?). If I remember correctly AMD implemented selective SiGe and Ni salicide on 65nm. Intel put these in at 90nm and if you look carefully at the yield/defect density plots, I think there was some irregularities on the 90nm curve. Both of these changes have yield and product sensitivities that need time to get ironed out - selective EPI is pretty pattern sensistive so there is significant work going from validating the process on test chips to working product and Ni salicide has always been delayed primarily for yield issues. It's a guess but I suspect one or both of these may have given AMD some fits early on in 65nm (which would show up as either yield or speed problems).

I tend to think 45nm (production-wise) should go relatively well for AMD.

However I do think some folks have lost sight of the forest staring at the 45nm trees when they talk about how much of an improvement it is.... the bar on 65nm was VERY low which makes 45nm look that much better. The clocks on 45nm are about what people expected on 65nm (with slightly better power #'s)

Anonymous said...

I would actually be surprised if they had major yield problems on 45nm - there is just not that much different from 65nm.

Admittedly I only saw one posting on this, from an Intel engineer over on Xtremecpus.com, so that's why I said "rumor" as in singular :). And that was several months ago. But I would think a relatively low volume compared to 65nm would show ramping problems after 6 months.

Isn't AMD supposed to issue revised guidance after Thanksgiving (Black Friday sales)? Their stock climbed all the way back to $2.35 before collapsing to $2.00 yesterday. If AMD does issue an honest revision, I wouldn't be surprised to see it drop below $1.50 - might be a good time to buy :).

Anonymous said...

Isn't AMD supposed to issue revised guidance after Thanksgiving (Black Friday sales)

At this point I don't think guidance matters (unless it is somehow better than the trend Intel set). Once Intel pre-announced the market likely priced in a similar fall in revenue for AMD. (That said from a market psychology point of view you may see a short term drop after a negative announcement - but I would expect it to rebound quickly if it was in line %-wise with what Intel pre-announced).

Full disclosure - I picked up some AMD shares recently... it's a question of whether you think the company is going bankrupt or not. While it could go a little lower, unless the company goes bankrupt this is pretty close to the bottom. I think with the cash infusion from the Middle East AMD will be solvent for a while - and once the foundry split is finalize (beg '09?), AMD will not have as severe a cash flow problem as that will be transferred to the foundry company.

My only concern would be if the company did a reverse split if it gets near the mid-low 1 range. If they did a 1:10 or 1:20 reverse split (essentially merging shares for those of you who are not familiar) that got the price back in the 10-20 range then I could see it dropping again. Unfortunately when the foundry split is finalized, there is a chance AMD could do something "funky" with the common shares like this.

Anonymous said...

http://translate.google.com/translate?u=http%3A%2F%2Fwww.hkepc.com%2F2044&sl=zh-CN&tl=en&hl=en&ie=UTF-8

AMD is planning 13 45nm quad processors between Jan and June... this would be in addition to the myriad of 65nm product bins. The 13 45nm SKU's cover a clockspeed spread of 500MHz... meaning MORE than 2 products on average for every 100MHz clock step!

Who the heck is doing product marketing at AMD these days?!?!?

You have quads under the Athlon 2 brand (the 0MB L3 cache models), you have Phenom 2 6xx, 8xx, and 9xx series all as quads and obviously the 7xx series will be tri-cores! At least they got away from the 5 digit names and cut it down to 3 of course they now have multiple brand names representing K10 quads which is always good for brand recognition!

Salesperson: Would you like an Athlon X4 or a Phenom 2?

Customer: I want one of the new K10 architecture quads.

Salesperson: So would you like an Athlon X4 or a Phenom 2?

Customer: I want the 45nm K10, not the 65nm one.

Salesperson: So would you like an Athlon X4 or a Phenom 2?

Customer: I want the 45nm K10 quad that I can plug in as an upgrade on my AM2+ board

Salesperson: So, would you like an Athlon X4 or a Phenom 2?

Customer: I'm confused

Salesperson: The Athlon one has no L3 cache.

Customer: Oh, then I want the Phenom 2

Salesperson: So do you want the 6 series, the 8 series or the 9 series?

Customer: Just give me an Intel chip!

SPARKS said...

LOL - Here they are by the numbers.

Phenom II X4 940 Black Edition - 3.0GHz
Phenom II X4 920 - 2.8GHz
Phenom II X4 925 - 2.8GHz
Phenom II X4 910 - 2.6GHz
Phenom II X4 810 - 2.6GHz
Phenom II X4 805 - 2.5GHz
Phenom II X3 720 - 2.8GHz
Phenom II X3 710 - 2.6GHz

Customer asks salesman, “Doesn’t Intel make a 920 and a 940, too?”

Salesman-“Sure they’re about the same performance wise, and the AMD chip costs less. We can pass those savings on to you.”

SPARKS

SPARKS said...

Oh yeah, let's not forget the Cripples.

Phenom II X3 600
Phenom II X2 200

Imagine that, they ridiculed INTC for "bolting" to dual cores together, but they are now selling half a functional die!

You can't make this stuff up.

SPARKS

SPARKS said...

Who the heck is doing product marketing at AMD these days?!?!?

You can't figure that out with that clockwork ticker?

They're selling garbage, of course.

SPARKS

Anonymous said...

LOL - Here they are by the numbers.

Phenom II X4 940 Black Edition - 3.0GHz
Phenom II X4 920 - 2.8GHz
Phenom II X4 925 - 2.8GHz
Phenom II X4 910 - 2.6GHz
Phenom II X4 810 - 2.6GHz
Phenom II X4 805 - 2.5GHz
Phenom II X3 720 - 2.8GHz
Phenom II X3 710 - 2.6GHz


You forgot the
Athlon x4 615 - 2.7GHz
Athlon x4 605 - 2.5GHz

I guess the tricore with L3 cache is better than a quad core with no cache? Or I guess it could just be another case of an F'd up marketing team?

There's also 'Athlon' version of the tricores (with no L3 cache) as a 4xx series. It seemed like the 2nd # was tied a specific clockspeed (x40 = 3Ghz, x20 = 2.8 GHz, x10 = 2.6GHz) , but then you have the Athlon X2 (45nm K10) which has a "240" at 2.8GHz.

The problem Sparks, is when you have garbage the product marketing becomes all the more important... trying to confuse the consumer with an endless amount of models for what is still a very small overall market (desktop quad) is crazy. While this segment will likely grow - desktop will continue to shrink and it just seems odd that AMD is throwing this many quad products at it.

I also find some of AMD's claims at analyst day inconsistent with their product roadmap - they claim they will be fully converted to 45nm around mid-year, yet 45nm dual core desktops are not due to launch until June... are they just going to have an instantaneous dual core crossover and stop selling 65nm dual cores in July?

I suspect the fully converted is quoted by wafer starts (which means add 3 months for product sales to be fully converted) and they are probably excluding any outsouricng (Chartered). This is the "world class conversion time" game that AMD plays and people like Dementia don't understand when they compare Intel and AMD tech node conversions. When Intel quotes crossover it is based on actually SALES, AMD fudges a bit and often uses wafer starts which makes things seem a bit quicker then they really are.

By the way - the NY subsidy was formally approved to be transferred over to the foundry company (a non-US chartered company)

Tonus said...

BTW, the TLB issue that Intel has mentioned in its instruction manuals-- is this the same issue that AMD had with Barcelona? And is it true that it exists in Core2 and Penryn processors?

Because if it is, then I don't understand why AMD chose the path it did when dealing with it.

SPARKS said...

“The problem Sparks, is when you have garbage the product marketing becomes all the more important... trying to confuse the consumer with an endless amount of models for what is still a very small overall market (desktop quad) is crazy.”

I don’t look it as a problem for AMD. I look at it as them propping themselves up as a big time manufacturer with a varied product lineup for investors and the industry as a whole. To those not nearly as informed us, it looks as if they’ve been successful, when in fact, they are selling as much as they can from a single wafer. As you said previously, “the bar was pretty low to begin with.”

They did the same thing with the initial Barcelona release (remember the 1.9 GHz SKU?), now that was GARBAGE. This is faster garbage. In fact you (correctly) made the same observation back then, but I suspect you think its worse this time, I concur.

In contrast, considering INTC’s product lineup, the choices are VERY real. Rocketing 65nM E8600’s or bread and butter Q6600’s at 189 bucks, with every point above and below.

Then 45nM LGA 775 enters the picture with a host of products from relatively cheap to the ridiculously obscene (My Chip, QX9770), and QX9775. THEY SOLD, despite the stupid prices!

Then there’s the future platform gorilla LGA 1366 with only 3 SKU’s from relatively cheap to moderately obscene (My future chip). With the 300 dollar 920, INTC is definitely sucking (those in the know) into a new high performance platform. DELL has introduced a Core I7 gaming rig for less than 1000 bucks!! Merry Chistmas!

http://www.pcmag.com/article2/0,2817,2334759,00.asp


Intel buyer’s (all) have real choices to make based on real performance based on cost for specific needs. The new platform introduction and the “old” platform overlap is perfect and brilliant.

I think when INTC starts to clear inventory as ROBO mentioned in this most recent post, AMD will need every bit of NYS money, Arab money, “creative accounting”, and marketing clowns to combat this onslaught waving over their heads like the Sword of Damocles. All the marketing clowns in AMD aren’t going to make things funny enough to keep this sword from dropping.

This will happen, as soon as we see a real Pheromone up for sale and AMD commit with real prices. INTC can afford to wait, and that’s exactly what they’re doing.

Yes, INTC made its choice a long time ago when we discussed it more than a year ago. They took the slow painful death (while making record #’s) as apposed to no profit and quick kill.

And I am still enjoying the daily daytime drama.

SPARKS

Tonus said...

Hrm. Based on an article in The Register, the TLB issue was first present in Core2, which is when Intel made the note in their programmer's manual. The problem was patched via BIOS updates prior to the introduction of Core i7.

I'm still wondering if AMD's issue was the same, and why they handled it so poorly. Based on Intel's actions, it could be a result of AMD's delays in getting Barcelona out the door. They had to warn OEMs and customers about the problem and schedule a fix, and the press ran with the story.

I guess the real question is, did Intel's BIOS patch impact performance? If not, why did AMD's?

Anonymous said...

Tonus said...
Hrm. Based on an article in The Register, the TLB issue was first present in Core2, which is when Intel made the note in their programmer's manual. The problem was patched via BIOS updates prior to the introduction of Core i7.

Yes - appears sombody inadvertently left this "errata" in the Ci7 manual, and of course the AMD fanbois jumped all over it just like they did with the Core2 TLB errata (which Intel fixed without any performance issues).


From the "ignorance is bliss" fanbois at UAEZone, somebody with a little common sense: I don't know what the truth is here, I'm neither an Intel nor an AMD fan-boy, I'm just trying to keep some balance because rushing down the "i7 has a TLB bug, let's shout about it" route might be flying totally in the face of facts and weaken the reputation of this board.

I really had to laugh at the last part :). Guess the poster doesn't realize that UAEZone's reputation is already about as low as, say, Sci' blog.

The only balanced & fair posters (enumae, Giant, carniver, etc) seem to get banned frequently when they stray outside the "AMD good, Intel evil" forum guidelines too far :)

Anonymous said...

At this point I don't think guidance matters (unless it is somehow better than the trend Intel set). Once Intel pre-announced the market likely priced in a similar fall in revenue for AMD. (That said from a market psychology point of view you may see a short term drop after a negative announcement - but I would expect it to rebound quickly if it was in line %-wise with what Intel pre-announced).

Yes, but... if the recession goes on for a couple of years as some are predicting, then UAE may cut their losses even though with oil under $50 a barrel their economy can still balance - I think only the Saudis are still economically viable if oil gets down into the $30 range for any length of time. The idiots in Russia, Venezuela and my personal favorite, Iran, were banking on oil being above $100 forever and spent or otherwise indebted themselves based on that in an effort to prevent the populace from throwing them out of power. Investing in new fab capacity makes no sense in a shrinking market anyway.

I think what happened here was that high oil contributed a lot to the current recession, but low oil by itself is not going to reverse conditions anytime soon. There's a lot of other problems that were exacerbated by the sliding economy that will need to be fixed as well now.

So if the above holds (and I know my predictive track record here is about that of a coin toss - still it's better than Sci's :) and since AMD most likely couldn't withstand a prolonged recession by itself, then their fate is almost entirely in the hands of the UAE investors. So the UAE is the one I'd be watching most closely, now that you have invested in AMD :). With the failed OPEC meeting just concluded, some are now predicting the demise of OPEC and hence dumping of oil by the above 3 idiotic governments in an effort to keep their economies afloat, which will mean even lower oil prices.

Anonymous said...

Tonus said...
BTW, the TLB issue that Intel has mentioned in its instruction manuals-- is this the same issue that AMD had with Barcelona? And is it true that it exists in Core2 and Penryn processors?

Because if it is, then I don't understand why AMD chose the path it did when dealing with it.


My understanding is that Intel wisely designed in the capability to revise the microcode via a BIOS update, or something like that. AMD however has to issue a new stepping to fix their problems or else turn off the TLB completely, which is why they suffered a 10%+ performance penalty on an already-low-performing CPU. Seems like Intel is the "smarter choice" :)

SPARKS said...

http://www.fudzilla.com/index.php?option=com_content&task=view&id=10742&Itemid=1


Well, Fuddie is saying Deneb will launching January. (I thought AMD said 4Q).

In any event, they missed another Christmas. That's three if your counting.

SPARKS

Anonymous said...

I'm still wondering if AMD's issue was the same, and why they handled it so poorly. Based on Intel's actions, it could be a result of AMD's delays in getting Barcelona out the door.

I think we had a long discussion on this a while back... I really don't think the TLB was ever a major issue for AMD - I continue to maintain they were using it as a convenient scapegoat for the problems they were having on power/clockspeeds (65nm process). They were struggling with 1.9-2.3GHz chips at the time, they (effectively) shut production down for 6months and then came out of it with a new stepping that not only had a TLB fix but also some better (relatively speaking!) clocks. I think the whole TLB fix was used to give them the time to work on the 65nm issues, in addition to fixing the issue.

Of course I have no evidence of this... and it is just speculation.

Anonymous said...

Anonymoose - you bring up some good points, but in a year the foundry will be a separate entity from AMD (and AMD stock). Yes AMD will own 44% of it so there will be some tie in with the AMD stock performance but in a year AMD will be buying wafers not having to invest in capacity.

I don't see UAE cutting their losses in the near term - most of their spends will be in 2009 for the F38 conversion buildout (and some additional capacity in F38). They are committed to this at this point. The next major commit is the NY fab capital, but those spends won't be until 2011/2012 (AMD claims NY fab online in 2012). So after the bulk of the capital spends next year, there is a relatively lull until 2011 - this could be effected by a prolonged downturn, but with Sparks 33% coupon on capital, it'll be hard to kill this (at worst it would probably be pushed out). The only way I don't see the NY fab being built out is is the foundry goes under.

SPARKS said...

Regarding the TLB ‘erratum’, I read somewhere you have these flaws in every chip ever produced. They can be small, easily addressable ones, or you can have your big nasty ‘erratum’ that difficult to address without serious performance hits, a la Pheromone I 10 % last year.

Is any of this true? Has there ever been a flawless CPU? (besides mine, naturally)

SPARKS

Anonymous said...

The TLB 'problem' is the web at its best. Some reader sent an email to Fudzilla with some info that had previously been posted on UAEzone.

Fudzilla, in true journalistic competency, saw the link to the erata document - felt no need to ask is this a big deal, research the issue or ask Intel or Mobo makers or server folk or ANYONE if this is real (though at least they acknowledged that the source was "a reader").

Since it was posted on Fudzilla, it therefor gets published on the INQ. This apparently is an actual site that some Intel folks must read as they provided a response to the INQ.

So basically some yahoo took the time to read through the erata document, made a bunch of assumptions on what was stated, posted it on UAEzone which then suddenly is a credible new story for Fuddia which makes it a credible story for the INQ.

So here's a couple of questions/observations:
- other than someone reading the document has their actually been a reported case of this erata occurring in the field or in a lab or during sampling or at Intel or... anywhere? (In the AMD case, while the erata was rare and obviously overblown, it did actually occur though was hard to reproduce)
- people (UAEzone) choose to ignore Intel's response that it was addressed in a previous BIOS prior to launch, since every single statement Intel has ever made has not turned out to be 100% true..so that clearly(?) means every statement made should be considered a lie. (Using this "logic", noone should actually believe what ant company or person ever says). Along these lines I'm going to throw it out there - Intel is actually lying about Core i7 being manufactured on 45nm... noone has actually measure these, they have lied in the past, so by default the whole Core i7 is made using 45nm must also be considered a lie by default! In fact the earnings #, the actual # of factories.... all lies!
- Even though many folks over at UAEzone acknowledge if this were an issue it would probably be a minor one - it doesn't matter as folks made a big deal of AMD's TLB issue. There are 2 obvious problems with this:
1) We have no way of knowing if the issues are similar at all (other than the both use "TLB" in the description)
2) It relies on the third grade logic of well our behavior/statements are OK, because so and so did it before. It doesn't matter if we are wrong as others were wrong before us... gotta love the intellectual honesty and simplicity of that logic!

Anonymous said...

This just in:

AMD today announced that it expects revenue from continuing operations for the fourth quarter ended December 27, 2008 to be approximately 25 percent lower than third quarter 2008 revenue of $1.585 billion, not including process technology license revenue. The decrease is due to weaker than expected demand across all geographies and businesses, particularly in the consumer market.

AMD will report fourth quarter 2008 results after market close on January 22, 2009. AMD will hold a conference call on this day for the financial community at 2:00 p.m. PT to discuss fourth quarter results. A real-time audio broadcast of the teleconference will be provided at www.amd.com. The webcast will be available for ten days after the conference call.


AMD's stock down slightly, by just 2 cents, but the above was released only about an hour ago so we'll see in the next few days how this is going to play out.

Anonymous said...

Is any of this true? Has there ever been a flawless CPU? (besides mine, naturally)

I seriously doubt that mere mortals could ever make an absolutely flawless CPU of today's complexity. Even if they could it would be prohibitively expensive. Back when I took "digital logic 101" in college we studied asynchronous "race" logic a bit, and quickly discovered why it is preferable from a design standpoint, for any logic more complex than a flip-flop uses clocks to synchronize them :). Well a clock signal is pretty simple - it only imparts two pieces of information - duration (length of time it is high or low) and sequence (a condition prior to a clock edge can be said to precede another condition that is only present after that particular clock edge). Now imagine all the possible permutations of all the input, output and intermediate (internal state) signals present in a modern CPU. It would likely be impossible to test it for all permutations to see if it met all design specs under all conditions, in any reasonable amount of time that is. So these deviations from the design spec get published and workarounds developed as they are discovered. If the manufacturer is lucky or has some foresight, an easy fix that doesn't affect performance can be made. Otherwise, they may have to take it off the market and wait for a respin :).

At least, that's how I see the errata process occurring :)

Tonus said...

AMD's updated Q4 outlook? Down 25%.

Intel's 14% expected decrease doesn't feel any less painful, but a 25% downturn... ouch.

Tonus said...

I just realized that the numbers are not comparing the same thing. Intel revised its expected Q4 revenue numbers down, whereas AMD is expecting 25% fewer revenue compared to Q3. There is normally a sales increase from Q3 to Q4 (holiday sales). It seems as if AMD decided to present their Q4 guidance in a manner that would seem the least negative.

Their revised estimate could be in the 30-35% range or worse, if they did it the way Intel did, wouldn't it?

Anonymous said...

Their revised estimate could be in the 30-35% range or worse, if they did it the way Intel did, wouldn't it?

That's my thinking as well. I suspect we will see AMD's quarterly results being a $500-600M loss again, unless they can come up with more imaginary numbers as well as sell more assets (and I don't think the Abu Dhabi deal will have gone through by Dec. 31).

Anonymous said...

Intel revised its expected Q4 revenue numbers down, whereas AMD is expecting 25% fewer revenue compared to Q3. There is normally a sales increase from Q3 to Q4...

While normally this assessment would be correct, AMD's previous guidance for Q4 revenue was flat, so it doesn't matter in this case whether the # is referencing Q4 guidance or Q3 actual as they are the same #. Of course they guided flat from the 1.5Bil as opposed to the 1.7Bil they reported because of the games being played with the tech revenue license fee they collected - they used that # to get favorable QoQ and YoY comparisons and artifically inflate the gross margin, but then conveniently took it out when guiding for Q4 (pretty sleazy)

unless they can come up with more imaginary numbers as well as sell more assets

I believe the CE division sale will hit the books in Q4 (I forget but I think this was in the 170Mil range)

The real question out of all this is that as AMD warned later than Intel, have conditions worsened more than when Intel warned (and thus Intel will see even more of a drop than they indicated)or is AMD just getting hurt more by the current conditions?

SPARKS said...

“I seriously doubt that mere mortals…………….”


Moose! Impressive, thanks I suspected as much.

CPU’s are like beautiful women, they’re not perfect under every circumstance, but we’ll live with the minor flaws if they’re fast and they do the right thing.

Just a little work around.

Great headroom.
And a nice cache.

SPARKS

Anonymous said...

Oil under 50 bucks

Arab Micro Devices biggest customer sales down 25%

TSMC employees going on forced long holidays

Wonder where the cash for them 50 million+ immersion steppers, million dollar plus A, B, C, D stepping masks are going to come from for poor green company...

SPARKS said...

"Sparks 33% coupon on capital,"

SPARKS

Anonymous said...

I'm wondering how done a deal the Luther Forest fab is? Don't the taxpayers have final say-so (via referendum, recall vote, etc. like the Californians do every so often)? I would think that if the recession is as long and as bad as the evidence suggests, somebody with some common sense in the state gov't will put the brakes on or cancel the deal entirely if not too late. An expensive, underused or mothballed fab doesn't bring in any jobs after the construction is done...

Anonymous said...

Don't the taxpayers have final say-so (via referendum, recall vote, etc. like the Californians do every so often)?

The taxpayers did have a say - they voted the yahoos in for years and continue to do so. The state is the classic democratic tax and spend philosophy. For them it is just a matter of increasing taxes; especially with a soon massive contraction in the tax base with all of the financial jobs being cut - not to mention the impact that will have on supporting businesses that rely on people with large disposable incomes. It is going to be a scary scenario as the tax the rich philosophy doesn't work when the rich base is shrinking rather quickly - soon it will be cut spending or tax everyone. Plan B is for an EU-like approach where Cuomo just starts fining all successful, I mean monopolistic, companies

This money isn't a referendum, or paid for by a bond issue or doing a special sales tax increase or something else - it is simply a use (misuse?) of state revenue via the budgeting process.

As the offer has been approved to transfer to the foundry company, I can only imagine what sort of financial penalty or what lawsuits might come from AMD (not that they would sue anyone?). I think the only way NY gets out of this Shoreham facility (inside joke), is if the foundry goes under. NY had a chance with the foundry spinoff - they probably could have wriggled off the hook by not approving the transfer of the offer, but they didn't.

On the bright side I have to assume the grant is based on matching investments - i.e if AMD doesn't build the fab out they won't get the entire payola. If NY had any sense it has to be based on a relative ratio to capital spends by the foundry (if the foundry only builds out 1/2 the fab I imagine they would only get ~1/2 the grant)

Anonymous said...

On the bright side I have to assume the grant is based on matching investments - i.e if AMD doesn't build the fab out they won't get the entire payola. If NY had any sense it has to be based on a relative ratio to capital spends by the foundry (if the foundry only builds out 1/2 the fab I imagine they would only get ~1/2 the grant)

Heh, well AMD has been producing only half a CPU for some time now, so this wouldn't be too far a stretch :)

Ed Stroglio and I have been exchanging comments on his blog and he assures me the Luther Forest is a "done deal", so I guess the NY taxpayers are on the hook for 1.2B or whatever the incentive was. I'll take Ed's word for it since he appears to have done a lot more research than the Lazy Moose :). He's also confident that $40 oil won't bother UAE for some time - I would have thought them to be more conservative but I guess not.

SPARKS said...

“NY gets out of this Shoreham facility (inside joke),"

I love it when you direct those unabashed comments towards me, it shows me you care.

By the way, your political commentary on how, “The state is the classic democratic tax and spend philosophy.", et al, is very much on the money. Forgive the pun.

I seriously doubt those outside the state can understand how we, the taxpayers, are bulldozed into catastrophic, state funded, fiascos. Typically, this is another excellent example of how the politicians throw good money after bad, especially after knocking down a sorely needed offshore natural gas facility, quite recently.

As you all know, my fondness for INTC is boundless. Honestly, the possibility of Arab Micro Devices going belly up so we (in NYS) don’t have to pay for the privilege of subsidizing their debt, would be a dream come true.

That said, if you can hear me “Big Paulie”, slash prices to the bone, and crush the cockroaches. Dumping 65nM inventory would make me weep for joy.

SPARKS

SPARKS said...

“Don't the taxpayers have final say-so (via referendum, recall vote, etc. like the Californians do every so often)?”

Ya know Moose, we are never lucky enough to get a good actor/politician as do the Californians. Take Ronald Regan (rest in peace), he as Governor turn the State on end when his ‘Welfare Reform’ included WORKING for your check be it a State fund job, or some other form of useful placement/arrangement. He went on to become one of the greatest Presidents of all time by technologically and financially bringing the Soviet Union to a collapse, hence, the end of cold war as we knew it. (Those who remember ‘duck and cover’, hiding underneath your desk, or hands over your head, face against the wall in the center corridor.)

Arnold Schwarzenegger actor turned politician became one tough SOB recently when he refused to sign the state budget until they cut out the pork and fat. He brought the State Legislators to their knees and exposed them for the exorbitant spenders they were. He couldn’t be bought (he has too much money plus he married a Kennedy) and he can’t be bribed. The man has fame but more importantly he has the interests of the State and its people in mind.

http://gov.ca.gov/

They even laughed, initially, at Jesse Ventura (Governor of Minnesota) another tough bastard who couldn’t be intimidated or bought off.

In New York City we are fortunate to have a very powerful and wealthy Mayor. You definitely can’t buy Mayor Bloomberg who works for a dollar a year. He is worth billions and he will be reelected for a newly mandated, unprecedented 3rd term. Sadly, the rest are career lawyers/politicians, with no business savvy and multitude of hidden agendas.

Our ex Governor was a confirmed whore master who walked away Scot free because of the dirt he had on everyone else in the state. Our carpet bagging Senator, soon to be Secretary of State got the best Senate seat a sloppy blow job could buy. Make no mistake; most politicians in NY are power brokers of the worst kind.

We will never see an AMD vote. As you surmised (correctly), the one silver lining in this terrible economy is that the bottom drops out of this deal.

SPARKS

Anonymous said...

Arnold Schwarzenegger actor turned politician became one tough SOB recently when he refused to sign the state budget until they cut out the pork and fat.

You know I have many liberal friends in CA who say "Actually, he's doing a pretty good job" (which is as good a compliment as they will give and it pains them to say it).

The problem with CA and what will be an issue across all states is the new pork barrel system of individual proposition/referendums - these mandate spending (by law) if they are passed REGARDLESS OF BUDGET CONDITIONS (except perhaps under extreme conditions). I think at one point CA, someone from that state can correct me if I'm wrong, was running massive deficits partly because they were forced to follow through on all of the propositions that passed and left no room for actual "normal" spending without running deficits.

There was a damn Zoo tax proposal on the OR ballot! You know what, I'm all for the zoo, but if it is so important, figure out a way to FIT IT INTO THE EXISTING BUDGET, don't just add a new tax! Cut something else if it is so important.

There is no thing such as compromise anymore - the only compromise these days is not what program needs to be cut to live within a state's means, but what program can I give to politician X to get politician's Y project done and who's tax do we jack up to pay for it.

Anonymous said...

Sparks said...

Ya know Moose, we are never lucky enough to get a good actor/politician as do the Californians.

Well I think the Governator is OK but I admire Reagan much more. As for Ventura, I'm just waiting for Hulk Hogan to get elected somewhere next :).

Maybe you should consider moving to a less liberal state :)

SPARKS said...

“Maybe you should consider moving to a less liberal state :)”

Cheese, Moose, thanks. I have always taken a bit of pride in not being another New York, left wing liberal son-of-a-bitch, honestly. Intrinsically, I cannot subscribe to the principle of throwing more money at social challenges. They don’t work and sometimes lead to rewarding failure by default.

Believe me, especially going to school in the Midwest (Jesuits); it was a tough rock to crawl out from under. All said there are many more like me, specifically those who want to work and aren’t looking for a goddamned handout.

Conversely, perhaps more relevantly, the same could be said in the corporate sector. The current day financial crisis/debacle is a fundamental example of rewarding people making exceptionally bad decisions at precisely the wrong time. Wrector Ruinz (and the board, “G”) typifies the corporate mindset shared by so many others in ‘Corporate Office’. Like so many politicians, they are simply removed from office only live VERY comfortably on the millions they squirreled away while in power. With the exception of blatant graft and corruption like Long Island County supervisors, Nassau and Suffolk, our illustrious Governor, they share many commonalities with corporate power brokers nationwide. Rarely is accountability, let alone punishment, exercised.

My love/hate relationship with New York resides on many levels, plus I’m paid well and I really know my shit. Working in the guts of the most famous buildings in the world, plus belonging to a brotherhood with the most talented people I have ever put my life on the line with, keeps me here day after day, year after year.

Taking these big old whores and renovation them to 21st century standards (and safety) is a challenge I would miss severely. Not only do I feel they need me, but I feel obligated professionally to do my part. Further, I would not be happy leaving without giving my small part in the WTC 2 reconstruction, even if I screw in one fucking light bulb.

The city gives me so much, I give it back.

I’m taking the kids in to see the “Christmas Spectacular” (again) next week. They look forward to see daddy in HIS big city, and enjoy some exotic food.

http://www.radiocity.com/events/christmas-spectacular-11-08.html

http://newyork.citysearch.com/profile/46315718
/new_york_ny/elettaria.html

A New York State of Mind.

SPARKS

SPARKS said...

Well there you have it. As I suspected (the above posts), Pheromone II’s are priced @ $235 for the 2.8 GHz and $275 for the 3.0 Black Edition. Obviously they are targeting the midrange 65nM Core2 quads and the lower end Yorkies. I suspect that’s precisely where their performance is. The INQ is reporting this despite the JAN 8 NDA. What happened to the 5+ GHz overclocking monsters?

I guarantee you; we will not see (credible) performance reviews until these things hit the shelves first. I still say a $300 920 Core i7 will blow Pheromone II out of the water. It’s simply no contest.

At the risk of being redundant, they will have to get past a fast Yorkie first, and it ain’t gonna happen. They are certainly not going to touch a QX9770, not even close.

SPARKS

Anonymous said...

Sparks said...

The city gives me so much, I give it back.

I’m taking the kids in to see the “Christmas Spectacular” (again) next week. They look forward to see daddy in HIS big city, and enjoy some exotic food.


Yes, I can see your point of view. I also see the high taxes, the traffic and the liberals running the place. Of course, I have about the same here in northern VA...

I live in the DC suburbs, about 25 miles from downtown, and I only make it in twice a year - usually to take visitors + family to go see the Smithsonian, White House, Mall or National Zoo. I'll probably go in this coming week to see the national Christmas tree, if this cold snap lifts a wee bit :).

I haven't been to NYC in over 20 years now, although I have gotten too familiar with the G. Washington bridge while driving up to Rhode Island to visit my daughter at URI. Kinda wierd looking over to see the Empire State Bldg but not the twin towers against the skyline.

When I eventually retire, I'm gonna move someplace warm and with a downtown riverfront - Jacksonville FL and San Antonio TX come to mind, since they are in income-tax-free states :)

SPARKS said...

"I also see the high taxes, the traffic and the liberals running the place."

Moose, I laughed like hell when I read that.

The traffic can be beat. Mass transit is the only way to get in, get around, and get out. The taxes, well you’re just plain screwed

Retiring to a nice warm place after retirement is a definite goal. Water, and lots of it, is a mandatory prerequisite.

SPARKS

InTheKnow said...

I read an interesting piece on how the financial markets got into the mess they are in now. Thought I would share it. For a fascinating, if very disturbing, read just click here.

Tonus said...

Good news for AMD: Phenom 4.4GHz OC in the wild. (Well, sorta. No clear word on cooling used, and no word at all on voltages.) Some word of mouth from review sites should help AMD, even if it's a small amount.

Bad news for AMD: Abu Dhabi gets new terms. And they're not favorable to AMD, compared to the previous deal. This is a bad sign, when you are forced to accept new terms so soon. If I'm not mistaken, the Abu Dhabi group gets more control than originally agreed to, while paying less than originally agreed to.

In other words, Abu Dhabi and AMD effectively agreed that AMD is worth a lot less than they thought it did just a couple of months ago.

Anonymous said...

Tonus: In other words, Abu Dhabi and AMD effectively agreed that AMD is worth a lot less than they thought it did just a couple of months ago.

And we all know how small a space there is between "worth less" and worthless :). I really wouldn't be surprised to see Abu Dhabi decide the whole deal is insane and pull the plug. Unfortunately for AMD, Abu Dhabi is their last and best hope for Phenomkind. So now I'm predicting a Sharikook "BK by 2Q09" except for AMD, not Intel.

Anonymous said...

And we all know how small a space there is between "worth less"

I think this essentially means their (AMD) equity stake has gone from ~44% to ~34%. With a minority interest, I still do not see how this does not constitute outsourcing of CPU production (which is only allowed up to 20% of total production under the x86 license).

I still don't know why Intel is not pursuing this for at least leverage for either the lawsuit or for the time when the x86 license comes up for renegotiation in 2010 (obviously there would be political implications of just pulling the license outright).

And I don't love the no double negatives!

Once these deals start though, there is generally a clause in the contract if the merger/investor pulls out - so generally the only realistic course of action is to tweak the terms like has happened in this case.

Anonymous said...

http://austin.bizjournals.com/austin/stories/2008/12/08/daily3.html?ana=yfcpc

Thank you sir, can I have another?!?!?

Apparently the 1.2Bil is not enough (oh and the throwins for waste treatment, infrastructure, that are slowly soaking taxpayers even more under the radar).

Well now apparently AMD is exempt from sales and local taxes for construction of the plant to the tune of another 27Mil.

Hey Sparks, when you put an extension on you house or re-wire, NY allows you to purchase materials without paying sales tax on the materials, right?

SPARKS said...

“Hey Sparks, when you put an extension on you house or re-wire, NY allows you to purchase materials without paying sales tax on the materials, right?”

As a mater of fact, I just re did the service on my home. An upgrade to 200A cost about $700 in materials, bypass meter pan, SQ D 42P, AWG 3/0, the works. At 8.5 percent sales tax it came to about 60 bucks. Hey, part of than is going to AMD, my tax dollars are definitely at work here! Hoo Ya!

It makes perfect sense that NYS taxpayers are paying more while the Abduls pay less! This is New York , man, you can make this stuff up!



When the deal was struck it was based on AMD’s value @ ~$5.00 a share. Its twenty day moving average has drop significantly since. The Abduls are only buying what is current market value. (They are not as stupid as AMD was when they bought ATI for 5.4B. The combined value of both companies was less than half of that by Jan. 2007!) DOOH!!!

I suspect they don’t want to layout 350M when the true value is around 150M, give or take. The renegotiated 35% stake, more significantly, as opposed to the previously mentioned 45%, is something entirely different.

“I gonna make ‘em offer they can’t refuse”- Don Corleone

This I believe this is commonly called a ‘bitch slap’. AMD’s gonna bend over deep and give a reach around to get this thing though.

“still don't know why Intel is not pursuing this for at least leverage for either the lawsuit or for the time when the x86 license comes up for renegotiation in 2010”

Intel has been VERY quiet on this. Strategically, I believe they are taking the line of least resistance----- now. However, based on these percentages when the Abduls purchase this “great investment opportunity for the future”, and the deal becomes finalized, INTC’s lawyers will be circling the wagons on what left of the "Scrappy Little Company”.

It don’t mean nothing until you sign on the bottom line.

SPARKS

Anonymous said...

If Q4 is down 35% for AMD what does Q1 and Q2 hold

Economy is still sinking

IT are still scaling back

No relief in sight

All AMD got is pricing to sell at every bigger loss that no one wants.

Prediction AMD stock will be below a buck by Q1

Tonus said...

anon: "I still don't know why Intel is not pursuing this for at least leverage for either the lawsuit or for the time when the x86 license comes up for renegotiation in 2010 (obviously there would be political implications of just pulling the license outright)."

Isn't this something that the companies would pursue as quietly as possible? I am assuming that Intel would send a letter to AMD's legal team and that AMD would send back a reply and a dialogue would follow. It doesn't sound like the kind of thing either company would want to do in the open (Intel might risk bad PR for seeming the bully, AMD might risk further erosion of share price if there is even a whiff of a chance of losing the x86 license).

Anonymous said...

Isn't this something that the companies would pursue as quietly as possible?

Normally yes, but I think it would send a message to Cuomo, et al (looking for another head to mount over the antitrust mantle on the way to federal office) that before they go off and look at Intel, the AMD deal and hence the whole NY fab "job creation" fiasco could be in jeopardy if the foundry company is a violation of the x86 license terms. Who knows it might even give the EU pause as AMD still does employ a few folks in Germany and probably pays some EU taxes from the Dresden fabs.

I don't car about the 50%/50% voting stake, a 34% equity stake doesn't constitute ownership of the company and thus this should be considering outsourcing of production - I mean the whole idea of the deal was to outsource the manufacturing so AMD wouldn't have the capital expenditures and simply do a pay as you go design house mentality!

It seems like this could/should be rather cut and dry and Intel could easily parlay this into an end to the AMD lawsuit, a reneg of the license to allow outsourcing, and in so doing may even create a little goodwill with the US (if not the EU) regulators.

Sparks - the terms are even more favorable as I'm pretty sure it is a 20day moving average now and the 20day moving average prior to deal closure and they use whichever is lowest. So if the share price continues to drop before deal closure, the Paula Abdul's are protected and pay even less!

Just wondering if/when the SEC/FTC/other US agency has looked at or will be looking at the deal. This deal has floating terms, is with a company that chose to incorporate in the Caman islands as oppose to the US (typically Delaware), and is farming out technology which is considered rather sensitive. There are significant export restrictions on technologies 65nm and better for US companies (this includes products, manufacturing equipment to make the wafers and I assume the IP?)

SPARKS said...

“the terms are even more favorable as I'm pretty sure it is a 20day moving average now and the 20day moving average prior to deal closure and they use whichever is lowest.”

There is a very interesting (complicated) dynamic going on here. I think I’ve got a handle on some of the variables, but some aren’t so clear. So, what have we got?

IP for one, but that’s clouded by whose IP. Is there a clear fault line on what infringes on INTC’s IP? I’m sure there are many cross licensing agreements shared between both companies, other than the x86 license.

The x86 license itself. Which clearly states under,“6.2. Termination for Cause.”



“the other party undergoes a Change of Control. For purposes of this Section 6.2(b)(7), "Change of Control shall mean a transaction or a series of related transactions in which (i) one or more related parties who did not previously own at least a fifty percent (50%) interest in a party to this Agreement obtain at least a fifty percent (50%) interest in such party, and, in the reasonable business judgment of the other party to this Agreement, such change in ownership will have a material effect on the other party's business, or (ii) a party acquires, by merger, acquisition of assets or otherwise, all or any portion of another legal entity such that either the assets or market value of such party after the close of such transaction are greater than one and one third (1 1/3) of the assets or market value of such party prior to such transaction.”

And, concerning Subsidiaries:

“if such entity does not have voting shares or other securities, at least fifty percent (50%) of the ownership interest that represents the right to make decisions for such entity and an interest sufficient to receive at least thirty percent (30%) of the profits and/or losses of such entity.”


http://contracts.corporate.findlaw.com/agreements
/amd/intel.license.2001.01.01.html

AMD’s still very much present debt. Sure, most of those notes are due 2012, nonetheless they are still a drain and when considering the 20 plus 20 moving average, what was a small interest payment previously is now a considerable percentage of AMD’s actual worth and operating income. They’ve paid out about ~ 170M, or better, per quarter.

Your very significant, “There are significant export restrictions on technologies 65nm and better for US companies”.

Cash- Currently (third quarter) 1.34B, the cash is dwindling to the tune of 200M per quarter. (This is right in line with keeping the creditors at bay).

The recent 25% revenue warning.

Stock price not moving, and it ain’t gonna move. Most of it is institutionally owned. That’s why it didn’t move when the above bad new was released last week. All the small time players dumped when the bottom dropped out of the market 6 weeks ago. Plus, the Abduls own the rest.

I don’t know about you guys, but my tea leaves are telling me this is FUGLY. Add the NYS variable to this sordid equation, no mater what ‘Sparks Coupons’ they get, it’s still FUGLY.

Oh, yes the INTC Licensing Agreement states they cannot reorganize under Bankruptcy protection. (to beat their creditors out of billions the way K-Mart did)

“the filing by the other party of a petition in bankruptcy or insolvency;”


SPARKS

Anonymous said...

if such entity does not have voting shares or other securities, at least fifty percent (50%) of the ownership interest that represents the right to make decisions for such entity and an interest sufficient to receive at least thirty percent (30%) of the profits and/or losses of such entity.

It'll be interesting as the Paulas' (TM) pour in new capital over time...AMD will have to match it now or they risk falling below the magic 30% #. There was mention of this in the original deal (when they were at 44%), where they had the option to put in more equity or see their stake reduced. Looks like they now have just ~5% to play with until they can try to get the license renegotiated (and you can bet AMD will fight to get rid of that clause!).

If I'm one of the Paulas', I'm not sure how I would feel about owning 65% of the company but having the Randy Jackson saying 'yo dawg I got the same power as you' - what happens if there is a conflict between other foundry business and AMD foundry business? What happens when they want to vote off Ruiz?

Tonus said...

That is something that I was wondering about. Now that they are two separate companies, what are the options for ATIC to simply buy out AMD's share and own the foundry company outright? I assume that there are clauses in the agreements to prevent this from being done on a whim. But with the obvious bargaining power that ATIC has, I am assuming that they've allowed themselves that option if circumstances are just right.

AMD and its fanatics have tried to spin the split as a positive thing, but there are (IMO) real dangers to them if they can't get their ship righted quickly. As this recent change in the contracts shows, ATIC will take steps to protect its investment, as any company would be expected to do.

And there's really no room to complain about the treatment they get from ATIC. That would be like a person who is dangling over the edge of a steep cliff, complaining that the lifeline that was thrown to him is too rough and itchy. Sometimes, you're in no position to quibble over the details.

SPARKS said...

Speaking of INTC’s Lawyers, they are on the offensive.


http://www.mercurynews.com/business/ci_11179803


SPARKS

Anonymous said...

Nothing earth shattering, but a decent read on Intel 32nm:

"Intel to extend high-k lead at IEDM"
http://www.eetimes.com/news/semi/showArticle.jhtml;jsessionid=1XDSAKVN3DSAKQSNDLOSKH0CJUNN2JVN?articleID=212300580

Key highlights:
- 32nm will use a 2nd gen high K solution
- Despite an endless parade of claims made by vendors, high-k/metal-gate technology is much harder to develop than previously thought. IBM's ''fab club'' is reportedly wrestling with the technology, while the foundries will not deploy the scheme until the 32- or 28-nm nodes. (quoted from article)
- 22nm likely to be double exposure immersion litho (EUV "probably won't be ready")
- A 'new' (read - tweaked) 45nm process has been developed for SOC, system on chip, designs (my guess here is they probably thicken the gate a bit and tweaked the transistor parameters to give back some transistor speed in exchange for further power improvements).
- no mention of trigate. I have to assume this is not happening on 32nm, or it would have been in the headlines for IEDM, and now may be iffy for 22nm? (22nm was my initial prediction)

Anonymous said...

INTEL finishes 2nd generation High K / Metal gate and IBM and the alliance are no where to be found... WTF?

SPARKS said...

‘G’- At the risk of unleashing your full wrath, Charlie over at the INQ has his spin on the latest INTC news. Ah,-----ok, now hold on, he mentions EUV. (yikes!)

Granted, he may have a few circuit breakers tripped and his knowledge is kind of limited, but he’s the kind of guy who puts his ears to the railroad tracks. Does he hear a train coming, or are they just the loose marbles in his head?

“The next step is EUV for 22nm. We are not supposed to have figured that out yet, but since they are openly posting jobs for EUV mask makers at the 22nm node, we kind of put 2 and 2 together to get 22.”

http://www.theinquirer.net/inquirer/news/961/1049961/intel-goes-for-immersion-at-32nm

Ok, I’ll bite, what’s the difference between a run of the mill genius mask maker and a EUV genius maskmaker?

(please don’t tell me about $20K a year)


SPARKS

Anonymous said...

Ok, I’ll bite, what’s the difference between a run of the mill genius mask maker and a EUV genius maskmaker?

Well one of the key differences is the type of mask. 193nm uses a transmissive mask... the light that goes through the mask is the light use to pattern the wafer. EUV uses reflective masks - the light reflected off the mask is the light used to pattern the wafer.

I'm by no means a mask guru, but my understanding is these masks are often many repeated layers of alternating materials which introduces also sorts of potential issues including the uniformity of each layer and defects (not to mention the cost of making the masks).

The other issue is the lifetime of the mask - the EUV is higher energy (shorter wavelength), and I believe that causes the mask to degrade faster.

Check out the Silicon strategies link above... Mark Bohr (Intel) even indicates EUV will probably not be ready for 22nm. Now I'm sure Charlie is a whiz at process technology, but I think I'll tend to believe a senior technologist at Intel. (Yeah I know I'm crazy that way!)

Recall this is the same Charlie who claimed Intel was having an issue with 45nm and was developing a 2nd 45nm process (this was well over a year ago). The one who completely botched IBM's gate first technology (He implied it meant they would implement the gate first and then on a later process revision get around to the high K!) and had no clue that gate first referred to the integration scheme. To put things in perspective, his knowledge in these areas is only marginally better than Scientia.

InTheKnow said...

- A 'new' (read - tweaked) 45nm process has been developed for SOC, system on chip, designs

This is actually going to be a big deal because Intel has yet to make a chipset on 45nm to the best of my knowledge.

That means they either have to

1) Use multiple gate depositions to put down SiON gates for the chipset transistors and then HK/MG transistors for the CPU.

or

2) Re-design the chipset to use HK/MG transistors. I'm pretty sure that the performance on the HK/MG transistors is sufficiently different to require a major redesign.

I can't see Intel "de-tuning" their HK/MG transistors for the chipset performance, but I guess that is another possibility.

In either case, it looks like a pretty big undertaking to me.

InTheKnow said...

- no mention of trigate. I have to assume this is not happening on 32nm, or it would have been in the headlines for IEDM, and now may be iffy for 22nm? (22nm was my initial prediction)

I would agree that if it was done on 32nm that Intel would be blowing that horn loudly at IEDM. But I wouldn't rule out 22nm yet.

That isn't supposed to hit the market for almost 3 years yet. Intel might think it is a bit early to tip their hand on 22nm just yet.

Unlike IBM et. al. who seem desperate to seem relevant at the cutting edge right now.

Tonus said...

From that EE Times article: "Intel is on track for 32-nm production readiness in Q4 2009."

It will be interesting to see where IBM/AMD stand at that point, in terms of processor development. Since it seems that they will skip HKMG until 32nm, is there a possibility that they will appear to be 'closing the gap' once again? Or maybe it won't matter by then.

Anonymous said...

It will be interesting to see where IBM/AMD stand at that point, in terms of processor development.

Oh I predict we will be spun like a top on 32nm. Recall AMD has a bulk Si (non-HiK process) they are developing for the foundry, which should be out ~6 months prior to the SOI high K and SOI bulk Si 32nm processes. I fully expect AMD to start trumpeting 32nm production with the foundry bulk Si, non HiK process in efforts to once again confuse the press into thinking they are 'closing the gap'. They have already started the propaganda campaign with the analyst day schedules which show bars for the process node starting a the point of FIRST SILICON! (which of course the UAEzone folks are comparing to Intel's actual product out timelines and I have seen a few "technical" websites also confuse with production)

32nm will matter for AMD as they should get a large jump from high K (assuming they do it right) while Intel's 32nm will be more of a typical node transition. It will just likely be filled with the typical "technically correct", highly misleading statements on process technology that IBM and AMD tend to put out.

Anonymous said...

Some more generic IEDM previews... but take a look at some of IBM's claims:

http://www.eetimes.com/news/semi/showArticle.jhtml;jsessionid=AYUBALJJNF3YGQSNDLOSKH0CJUNN2JVN?articleID=212400168&pgno=2

Taking a longer view of the semiconductor roadmap, IBM will announce at IEDM what it claims in the world's smallest reported SRAM cell size: 0.1 microns2

Well, technically true... but Intel's 32nm cell size is 0.171um2, so 0.1um2 on the next generation is not that great (since SRAM should achieve 50% scaling). It is likely the world's smallest because Intel hasn't reported their values yet! 50% scaling for Intel would put them at ~0.085um2 (or 15% smaller than IBM). Also not mentioned (and I'm not certain of) is if Intel's claims are based on an 8 transistor cell (Intel added 2 transistors to the SRAM cell for improved power management) and it is not clear if IBM is using a 6 or 8 transistor cell for their claim.

What is also a bit odd? IBM's "22nm" SRAM cell uses 25nm gates! I wonder if they are printing anything at 22nm or smaller or are just calling it the 22nm node.

They also claim a single metal gate material with their high K process, but..... 2 different "interface layers" to tune the workfunction (for NMOS and PMOS). This makes it sound like a single metal solution but in reality is no different than what Intel is doing (with obvious exception of gate first vs replacement flow)

SPARKS said...

“EUV uses reflective masks”….. Naturally, I had no clue.

Sooo……… I did a little ‘light’ reading. (sorry) The number of patents on the technology out numbers the amount still born cripples on an AMD wafer. I got these links for both AMD and INTC below.

The tools (read: lenses), the mirrors that reflect filter out the less desirable (longer) ‘Reflective Spectral Filtering’ (Christ) wavelengths, and the number of layers necessary, seems to be extremely complicated and expensive. Seems like a lot of places to introduce distortion and error, multiple times over.
Couple this with extremely powerful EUV that eats the tools and a great deal of energy lost by the time they hit the substrate, and you’re in for a world of hurt if you don’t have this thing dialed in just right.

The thing that got me was the complicated lenses and reflective surfaces that get hit before you get home. Now I can understand your apprehension concerning the tech, not to mention the cost of these bastards. Whew.

These are the links, and if anyone’s inclined, there nice drawings of interest, if you’re verbally challenged (like me).

Interestingly enough, AMD does have the earliest patent and quite a few of them, to boot. IBM surprisingly doesn’t have that many. Of course, I favor the INTC patents.

http://www.google.com/patents?id=pdgFAAAAEBAJ&dq=reflective+mask

http://www.google.com/patents?id=jVIMAAAAEBAJ&printsec=
abstract&zoom=4&dq=reflective+mask

http://www.google.com/patents?id=Ai4QAAAAEBAJ&dq=reflective+mask

These are just a few. They’re gonna need a lot of lawyers to sort this all out when (if) this tech comes to fruition.

SPARKS

Anonymous said...

The thing that got me was the complicated lenses and reflective surfaces that get hit before you get home

Also consider that you have some light (EUV) loss at every reflection - this leads to some of the power demands for the WUV source (which in the recent past was still at least 10X from what was considered manufacturable). If you can't get enough light to the wafer then you need to increase dose time which means lower tool throughput (and I imagine introduces other process issues as well?)

And this is where claims of printing a feature or a wafer (like IBM did) is somewhat misleading - the process used may not be manufacturable at all and may require a complete rework to make it manufacturable.

Tonus said...

A humorous aside: This topic at AMDZone. A 'new' member posts a question regarding whether or not AMD could close the process gap with Intel by having TSMC manufacture its CPUs at 40nm. The first reply is from one of the more deranged posters there, giving him a dressing-down and implying that he's probably a pro-Intel troll, and if he's just willing to sip some kool-aid he might be saved.

The 'newbies' reply is priceless.

Tonus said...

Whoops! The second reply is the one from the nutcase. The first reply is short and straightforward.

Anonymous said...

^ Heh, I saw that too this morning. Looks like Zhooty-zoot-zoot got his nose in a toot.

BTW, I think he also posts on Toms as SighO2 or something equally lame. Although he may have been given a temp ban there.

Good ol' UAEZone - good for a laugh if not much else. So that's why I hope AMD doesn't go belly-up anytime soon - the fanbois would disappear too.

Haven' seen Sci update his blog in 4-5 months now - maybe he forgot he has one :).

Anonymous said...

Scientia is lost as there is nothing even a blind devoted fanboi can croon about these days

SPARKS said...

“Scientia is lost as there is nothing even a blind devoted fanboi can croon about these days”


‘Performance per watt’, blah, blah, blah.
‘Pheromone II, overclocks to 6 GHz’, blah, blah, blah.
Direct Socket Replacement, blah, blah, blah.

I decided to check the ASUS website for upgraded drivers and/or BIOS when I found this little nugget.

http://usa.asus.com/products.aspx?l1=3&l2=179

Motherboards that are specifically designed to handle 140W CPU’s!?!?! These were all AMD boards! Why? Do you think they might FRY older motherboards, hmmm????

I can see it now. Little Johnny Overclocker blows $275 on a Pheromone II ‘drop in replacement’, and presto, more smoke than mirrors.

It’s just something to think about, when considering the drop in replacement to an AM+2 platform at the end of its (long overdue) cycle, or swing over to fresh Core i7 platform at the beginning of its cycle.

As for me, you guys know I wouldn’t be in the same room with an AMD platform. I’m a poster boy for the INTC lunatic fringe but not as bad as that zooty dude, cheese. And, what the hell is spintel? Spin, he says? I got dizzy the way AMD spun that mess called Barcelona since the beginning of 2007! I don’t know what the color of the sun is on that guy’s planet but he should realize that AMD is merely a shell of the company it once was. Technologically and financially, they have been blown out of the water.

Wrector Ruinz, the terrible.
Thing are looking good, we’re only losing a billion a quarter!
ATI, The platform that never happened.
Triple Cripples!
Whoops, we forgot the dual core market!
Oh where, oh where, has my KUMA gone, oh where, oh where could it be? Woof!
Have you seen a competitive AMD notebook anywhere?
Sweet little Atom, doesn’t anyone want to fight you?
Arab Micro Devices
Duo’s out of Quads, nah, things can’t be that bad, can they?
What’s the mater, can’t you handle DDR3?
Sure you can do 32nM, right, tomorrow.
EUV? More like Lep-Ro-Sy

God, I can’t stand the way INTC spins this stuff!

“The Perfect Storm”, is getting more perfect every day.


SPARKS

Anonymous said...

Where is Hector these days?

You remember when you was challenging Paul to a benchmark duel?

Hector who is laughing all the way to the bank and the benchmark races these days?

SPARKS said...

“Hector who is laughing all the way to the bank and the benchmark races these days?”

Yes, laughing all the way to the bank, true, very true. However, I must beg your pardon; I believe “the benchmark races” comment is not entirely accurate. You see, he will be laughing all the way to the Saratoga HORSE races with money funded by the New York State tax payers! And, we all know how much Wrector loves to gamble, don’t we?

HOO YA!

SPARKS

Tonus said...

I saw this report linked at Aces. The person who posted it quoted this line: "Intel is saying that its 32 nm technology has the highest drive currents reported to date for NMOS (1.55 mA/µm) and PMOS (1.21 mA/µm) transistors."

I don't understand what they're talking about, but was hoping someone could explain (in layman's terms!) what the benefits are from having 'high drive currents for NMOS and PMOS'. Does this mean less leakage, and therefore higher clocks at lower voltages? Or is there a different (or additional) benefits?

Anonymous said...

I don't understand what they're talking about, but was hoping someone could explain (in layman's terms!) what the benefits are from having 'high drive currents for NMOS and PMOS'.

I'm no expert but I think the drive current is the 'useful' current available at the output node. Leakage current (through the gate or through the channel when the transistor is supposed to be off) is waste. So higher drive currents mean higher efficiency if the leakage currents are the same.

For a given fanout, higher drive currents also mean faster switching speeds. So what Intel is bragging about here is that they have fast and presumably efficient transistors at the 32-nm node.

Anonymous said...

Hexus has interesting comments on the "new" dual core Kuma (65nm, K10 based).

http://www.hexus.net/content/item.php?item=16614&page=2

Robo - I think you need to dig up your picture of a tri-core and update it (though Hexus' cartoon in the link was funny too). I guess we should call this a 'non-native' dual core?

Do you remember the olden days of the tri-core discussion and many fanboys were saying how speed bins should be much better with one core off? Well with 2 cores off, AMD has managed to eke out 100MHz over the top quad core bin! (And at a mere 95 Watts!)

The new dual-core chips won't be called Phenoms, strangely, because that will probably be viewed as dilution of the performance brand.

"Dilute" the phenom performance brand? Nearly fell out of my chair when I read that!

It's clear that the new Athlons' K10 architecture, based on Phenom quad-core, is better than the K8 Athlons that have held the fort for many years, but the comparative lack of clock-speed (at a given price, say £65) means that the value proposition isn't that different from AMD - for the same cash you can buy 'x' amount of performance; it's just the architecture that has changed.

So this begs the question - why release this on 65nm with 45nm around the corner and it being bigger than dual core K8's (and AMD probably not being able to get much premium for it)?

1) Are these just chips with 2 bad cores? Why intentionally disable 2 good cores when it is a lot bigger/more costly to manufacture then a K8?
2) Is this the plan to clear out quad core and tricore 65nm inventory as 45nm ramps up? (i.e don't want to slash 65nm quad pricing and deflate quadcore sales - so make them into dual cores to force those wanting quads to 45nm?)

I really don't understand this maneuver with 45nm dual core due at midyear, unless the 65nm yields are truly horrendous.

Oh and by the way it appears these don't OC worth a damn. AMDzone (which is not one to shine a negative light on AMD products!) could not even get this thing stable at 3.3GHz with 1.5V applied (the part is a stock 2.7). Funny how they said they'd imagine a 3.1 or 3.2 is more reasonable but didn't even test it! Since 3.3GHz wasn't stable at 1.5V, they just assume 3.2 would be? Could you imagine the outcry of "bias" "shoddy/incompetent testing" and "paid off rebiew site" if similar testing and conclusions were done to an Intel chip?

SPARKS said...

This is rich.

First, Triple Cripples

Now, Double Troubles

What’s next Single Slobs?

I told you they were selling garbage.

These were all rejects INTC would have called scrap. AMD saved all the duds and waited till they had enough to market as a dual core.

“that’s the way to do it,
money for nothin,
and the chips for free”

I want my,
I want my,
I want my AMD.

Money for nothin!

SPARKS

Anonymous said...

I want my,
I need my,
I need my UAE!

Money for nothing, and your fabs for free!
(well...at least 30% off)

Anonymous said...

Sparks said: What’s next Single Slobs?

Personally I vote for Single Dingles :)

SPARKS said...

Oh, oh, the perfect storm gets more perfect.

http://channel.hexus.net/content/item.php?item=16642

SPARKS

SPARKS said...

And, this.....(ITK, TONUS, fill us in on this)

http://anandtech.com/weblog/showpost.aspx?i=532


SPARKS

InTheKnow said...

ITK, TONUS, fill us in on this

All I can say is that if it looks too good to be real, it probably is. I'm not trying to say that somebody cheated the benchmarks, I'm just saying there is more to this than a simple number reveals. My suspicion is that SAP is a benchmark that favors the Nehalem architecture in some way.

Anonymous said...

Why is high drive current good?

Well drive current is the measurement of the strength of your transistor for a given leakage. You see lots of FUD from others.

Higher drive for a given leakage means you get faster transistors which means faster clock frequency for a giving leakage. Of course faster frequency also increase total power because of the faster switching. But a good measure of how good your technology is is the drive current for a given off leakage. Notice how IBM/AMD don't disclose anythign about 45nm or 32nm but talk about something INTEL hasn't published yet. INTEL for the past few generation has had much higher drive current then any other company. AMD/IBM try to avoid direct comparison by talking about 22nm but if they results of the past 5 generations are of any indication I think when INTEL publishes 22nm it will far surpass the men in green.

Tonus said...

ITK: "My suspicion is that SAP is a benchmark that favors the Nehalem architecture in some way."

Or perhaps the other way around? Nehalem is designed to be a competitor in the server space, would it be possible that Intel optimized the platform for high end DB and similar apps? According to Johan, "this is a certified by SAP benchmark". He is suspicious of the performance improvement, though.

It will be interesting to see how this shakes out. It'd be pretty embarrassing for this to be either a scam benchmark or one that was performed so sloppily that it amounts to a scam. How would SAP look if they certified a benchmark that wasn't kosher?

InTheKnow said...

I decided to wander over to AMDZone to see what the AMD faithful made of the SAP scores. I didn't find anything about that, but I did see Scientia proclaiming his interpretation of DFM once again. He continues to labor under the mistaken impression that this is something NEW!!! and that AMD just discovered it for 45nm and this discovery has eliminated an Intel manufacturing advantage.

So let's be real clear, DFM is a process that all manufacturers use to deal with process variability. I don't care if you are making mousetraps, you use some level of DFM. The stapling machine that mounts the spring mechanism on the board has some variance and the board the spring mechanism is stapled to is made large enough to accommodate that variability. That is DFM in a nutshell.

Now if you really want to get into the gory details of how to apply this for semiconductors all you have to do is read this paper by Intel's Kelin Kuhn. It isn't light reading, but it will give you a good idea of the types of things that are accounted for by DFM. As you can see, this isn't something new, nor is it something that you can just turn on and Voila! you are in business.

InTheKnow said...

Or perhaps the other way around? Nehalem is designed to be a competitor in the server space, would it be possible that Intel optimized the platform for high end DB and similar apps?

Sure it is possible, but I don't think it is likely. Nehalem's basic architecture has to accommodate server, desktop and notebook products. That would tend to discourage product specific optimizations.

And as I said, I don't think that the benchmark is bogus. I'm just not sure that all the details of the configuration that we are looking at are comparable. There may be something in the processor or the platform that makes it particularly good at this benchmark.

Or it could just be that the AMD crowd was right all along and HT with IMT really was the magic sauce for K8. Now that Intel has the magic sauce, we can really compare the processors in an apples-to-apples fashion. And Intel doesn't seem to be coming off to badly at the moment.

Anonymous said...

I didn't find anything about that, but I did see Scientia proclaiming his interpretation of DFM once again.

It is rather comical watching Scientia and Abinstein absolutely butcher process technology and manufacturing and then see 2-3 replies with "thanks" and "excellent analysis"

There is a post on 45nm with someone asking if a metal gate could be used with SiON (which is a good question).

Abinstein replied SOI and metal gate are not mutually exclusively... apparently he thinks SiON (the gate oxide) refers to SOI? Or he just chose to answer a different question that wasn't asked...

And then Scientia chimes in "remembering" something (of course there is no link or actual support which he would hammer someone else for doing) and says "I couldn't see how AMD could do any better with 45nm" This of course would kind of imply AMD has dropped CTI on 45nm... and tomorrow he'll be praising the CTI approach again. If you believe AMD and their CTI spin... the process should be continuously getting better for the next 2 years.

Then you have Abinstein completely misinterperting IBM's "22nm" FINFET IEDM paper. He doesn't mention that the 22nm cell is actually substantially LARGER than Intel's 32nm cell (though in fairness this is because it is a FINFET based cell). He doesn't know the cell was made with e-BEAM lithography (purely a research tool at this point with ridiculously low throughputs) and the HK/MG gate stack used is not a performance based solution, but one simply used for eval purposes... He also apparently has concluded that IBM was the first to make a 22nm SRAM cell and doesn't draw the distinction that IBM was the first to ANNOUNCE a cell (something I don't think Intel typically does until they have a fully functional SRAM chip). There was also a GLARING lack of any actual transistor performance data in IBM's announcement (again probably because they used a basic high K stack for research purposes and didn't integrate a performance solution)

The "22nm" comes into play as the cell 'demonstrated' is 0.126um2. Then you read the fineprint in the announcement: In simulations of SRAM cells of 0.063μm2 area, equivalent to or beyond the cell scaling for the 22nm node the results confirmed that the FinFET SRAM cell is expected to offer a significant advantage in stable operation compared to a planar-FET SRAM cell at this generation.

So basically they have built a 22nm cell (using a non-production worthy litho process) which is 2X in size of what a normal SRAM cell would be and used SIMULATIONS to conclude that it is expected to be better. One would think they would actually build a 22nm planar cell and not trust things to a simulation.

So all they have to do now is produce this with a manufacturing worthy litho process (immersion or EUV), use a high performance gate stack and front end process, and build a conventional 22nm cell to see if it really is better or just simulated to be better. Oh and instead of that whole "scaling thing", your 22nm SRAM cells will be larger than 32nm. There was also no mention of compatibility of the cell with the various strain technologies (this could either be anything from a nonissue or a total nightmare). SWEET!

The announcement is not completely useless, building a FINFET SRAM cell is no simple task and a significant achievement, but I don't think folks quite understand what the announcement actually represents.

Tonus said...

That's why I've learned to just look at this stuff more in hindsight than in a predictive manner. There are a lot of armchair experts who know just enough to be wrong.

Thanks ITK and Guru for the replies and explanations.

Tonus said...

"That's why I've learned to just look at this stuff more in hindsight than in a predictive manner. There are a lot of armchair experts who know just enough to be wrong."

A good example of this appears to be the new Kuma processors. Poor performance, high power consumption. Ed Stroligo linked to a review (which he felt was representative of the many Kuma reviews now out) which implied that with Kuma, AMD is effectively competing with itself because Intel's pricing and AMD's lack of competition at the high-end have compressed their product line. The reviewer seemed almost broken hearted over it, which is kind of funny (and which won't save him from angry cries of "PAID OFF FANBOI" from the raging hordes).

If the SAP scores hold up and show the real potential of Nehalem on server workloads, it would be another example of why more people should be quiet and let the GURUs of the world explain things. :)

Tonus said...

ITK: "I decided to wander over to AMDZone to see what the AMD faithful made of the SAP scores. I didn't find anything about that"

I think they discovered the chart that Johan posted, and they appear to think that it is based on benchmarks run by Anand himself. Thus you get the usual litany of accusations of poorly-run and biased scores. They're even accusing him of incorrectly reporting whether or not HT was used, even though these are benchmarks from SAP's own site, not Anand.

SPARKS said...

Cheeze, thanks fellas, but what the hell is a SAP score, and the doubts about the benchmark run?

Why is it too good to be true?

Nothing is out of my beloved INTC's reach these days.

Hoo Ya!

Tonus said...

SAP makes Enterprise Planning Software (ERP). I see ERP as a really large managed database that ties a companies' various departments together in order to better manage and track resources (expenses, people, etc).

Or to let answers.com tell it: "SAP stands for Systems Applications and Products in Data Processing. It is divided into modules that include Financial Accounting (FI), Controlling (CO), Production Planning (PP), Materials Management (MM), Sales and Distribution (SD), etc."

Tonus said...

Argh. ERP = Enterprise Resource Planning. Bit it is software. :p

Tonus said...

Double-argh. Bit = But!

SPARKS said...

Okaaay, I figured it was software.

What I don't understand is why you guys don't believe Nehalem did so well?

Considering the performance difference between Core2 desktop components and Opteron, why is so hard to believe that this new archectecture scales the same way on the HPC front.

Actually, weren't we expecting Core2 on steriods?

SPARKS

InTheKnow said...

Sparks asked...
why is so hard to believe that this new archectecture scales the same way on the HPC front.

This really isn't about scaling on the HPC system level. We are only looking at a 2P system here. What I find surprising is that Nehalem is over 100% faster than Penryn with a 10% higher clock.

Core2 was something like 30-40% faster than P4 IIRC. 100% is high enough to make me look for the catch.

But the more I look, the more I come to the conclusion that it really is all about the IMC and QP. Intel has been saying for a long time that the FSB was good enough for the desktop and that seems to be true for most apps. It seems that what they weren't saying was how woefully inadequate it was on server systems.

The AMD fans seem to have been right on that one. I wonder how they are going to like it now that Intel has "copied" AMD and they have to play on a level playing field.

InTheKnow said...

Intel's Kelin Kuhn (who I believe is in charge of transistor development for 22nm) doesn't seem too positive on FinFet's for the 22nm node. She didn't rule it out, but did put it in the "high risk" category.

Tonus said...

sparks: "What I don't understand is why you guys don't believe Nehalem did so well?"

The performance leap was something like 119%, if I am not mistaken. That's a pretty large performance leap. I was impressed by it, but not sure what to think since it's not an area I know much about. The consensus amongst more informed people seems to be what ITK said above, that the IMC/QPI and HT had a big effect there. Over at Ace's they were saying that SAP software is ideal for HT.

I admit I was one of those people who curled his nose at the thought of HT making a return, but this is why I'm not designing CPUs for intel. :)

SPARKS said...

“seems that what they weren't saying was how woefully inadequate it was on server systems.”

No, market demand was. So was everyone else at the time. The hand writing was on the wall. It all began with this:

http://www.geek.com/articles/chips/amd-releases-opteron-its-hammer-time-20030422/

I remind you of discussions like these:

http://computing-intensive.blogspot.com/2006/07/imc-myth.html


Now it is very (and painfully) obvious that Nehalem is going to live up to AMD’s Barcelona “performance simulations” back in 2006-2007, and some.

Was it truly “Woefully inadequate” or did INTC do ‘Barcelona’ right, when the time was right, like a smaller die, a superior process, a superior architecture, QPI, DDR3, superior branch prediction, extensive R+D, HI-K, etc.

ITK, Tonus, I respectfully disagree, (I’ve developed quite a bit of respect for you guys over the years) but there’s no magic sauce here, INTC’s engineers broke their collective asses for years to get this thing right, and it has all come to fruition. What I suspect is that many folks don’t believe it’s this good.

“The performance leap was something like 119%, if I am not mistaken. That's a pretty large performance leap.”

Keep the faith; no one is ever going to put huge billboards on the Hemsley Building on Park Ave. ridiculing INTC again. No more company picnic sky writing fly bys.

I don’t know where LEX is but the hand writing is on the wall. I’m loathing saying this but:

“Tic Toc, the clock is ticking.”

SPARKS

Anonymous said...

Tonus said: I admit I was one of those people who curled his nose at the thought of HT making a return, but this is why I'm not designing CPUs for intel. :)

Me too :). But I can understand Intel's philosophy here - Turboboost for the single-threaded apps; HT for the multi-threaded apps. The question I have is whether or not HT is automatically enabled or disabled like Turbo mode, depending on conditions. As Abinstein said, it would be a problem if you had to reboot to enable/disable HT.

InTheKnow said...

Was it truly “Woefully inadequate” or did INTC do ‘Barcelona’ right, when the time was right, like a smaller die, a superior process, a superior architecture, QPI, DDR3, superior branch prediction, extensive R+D, HI-K, etc.

Don't get me wrong, Intel did the job right with Nehalem. I fully believe that 45nm made Nehalem possible with the advent of HK/MG. Doing it before that would have been too big and too power hungry.

Where they blew it was in underestimating the benefits of Hyper-transport in the server space. They left the door open for AMD to make a name for themselves for far too long. And they have known this for several years now.

And I don't buy the "right time" argument from Intel for QP anymore than I buy the "right time" argument from AMD regarding the 300mm transition. Those are the type of changes you make as early as you can because it keeps you competitive. Intel could have done QP on an earlier process node just as easily as AMD did, but they were a bit full of themselves, imo, just like they were with P4. And Intel is still recovering from that strategic misstep, as they can't offer a competitive product in the 4P and up space.

All that said, Intel has clearly pulled their head out and has AMD on the ropes at this point.

It is like I said quite some time back. AMD really blew it when Hector started in with his dual core challenge stunts. Publicly mocking Intel made it personal for Intel management and that may have been a fatal error.

Anonymous said...

I just hope Intel pushes Westmere up earlier than a year from now. I might just have enough patience to hold off upgrading my ancient QX6700, but it's gonna be close! :)

Since AMD is pushing their 32nm out to 2011, does this mean they will be 2 years behind Intel, or just a little over 1 year?

Frankly I'm a bit surprised at the early Deneb reports (not the OC shenanigans staged by AMD). Hopefully this won't be AMD's dying star. I do find it "peculiar" that the official release date is Jan. 8 and yet none will be available until after the Q4 earnings call on Jan. 22nd - guess they gotta have something good to try and divert everybody's attention from what is sure to be a humongous loss...

Anonymous said...

Hmm, light sweet crude declined to close at $36.22 today, despite yesterday's OPEC production cut announcement. If this level is sustained, only the Saudis will still have their budget in the black. I gotta believe the UAE is rethinking their AMD deal again...

InTheKnow said...

RANT WARNING ON

Sparks, on an unrelated note, it is things like this that give unions a bad name in this country.

...a "jobs bank" program — negotiated by the United Auto Workers and the companies — under which laid-off workers receive unemployment benefits and supplemental pay from their companies for 48 weeks. If they remain laid off beyond that, they move to a jobs bank in which the company provides about 95% of their pay and benefits.

If that deal isn't socialism disguised as a contract, I don't know what is. No free enterprise or incentive to work there, my friend.

The automakers are as much to blame for agreeing to such nonsense as the unions, but this kind of cradle to grave job security is something the rest of us don't have. This is typical of unions 1-sided negotiating philosophy (either real or perceived) that pushes jobs overseas. This clause in the contract results in little sympathy for union workers and has got to be killing the US automakers.

Having vented, I will say that over the years my position on unions has moderated somewhat. My father-in-law is a member of the IBEW (your union I believe) and I have seen some of the good things a union can provide. But things like this travesty above give all unions a bad name by association and have just got to go.

SPARKS said...

“But things like this travesty above give all unions a bad name by association and have just got to go.”

ITK, I am in total agreement with you. As far as I’m concerned this is a political bone in which the Fed is throwing to labor/management with our tax dollars. Not only are they funding the Auto Industries recovery, they are funding labor until they can both collectively get back on their feet.

You mentioned the IBEW. We have no such entitlements, and never will. Obviously, the government feels the UAW workers are very important to the health of the nation on the whole. I can assure you this is an historic precedent in which we will never be privy to. (That because we are in the private sector). This means the auto in its entirety is nationalized by default

In our union, when there are a lot of guys out of work, we are mandated by the Union to take 10 weeks furlough to allow those guys to work. We get unemployment and a supplemental income check during that period (money put aside during the 42 week work cycle). The total comes to about 85% of our net. Basically, we take care of ourselves/own.

Perhaps they didn’t put the money aside because of perceived job security. Who knows?

Well, there it is, job security complements of Uncle Sam. The UAW is very special, indeed.

SPARKS

Anonymous said...

Looks like AMD has shafted their stockholders once again. From http://news.moneycentral.msn.com/ticker/article.aspx?Feed=AP&Date=20081219&ID=9467369&Symbol=AMD:

AMD cancels debt deal with Lehman Brothers

Advanced Micro Devices Inc. delivered another dose of bad news late Friday to shareholders who have already endured a brutal three-year slide in the company's stock price.

AMD, the world's No. 2 microprocessor maker, said the bankruptcy of Lehman Brothers Inc. forced the company to unravel a complicated deal with the investment house that would have protected AMD shareholders from dilution when buyers of $2.2 billion in AMD debt eventually try to cash in their convertible shares.

The deal was supposed to work like this: AMD paid Lehman Brothers $182 million to scoop up enough AMD shares before 2015, when the convertible senior notes come due, so that it could hand those shares back to AMD and cancel out the new shares that would flood the market when the senior notes were converted into shares of stock.

AMD, like other Lehman creditors, is trying to recover its money. AMD still has options, though, for reducing dilution from the debt offering, including buying back shares on its own. Now might be a good time: AMD shares are trading at lows they haven't seen since the early 1990s.


So even though the stock went up 4 cents yesterday, I bet it's gonna tank on Monday.. And I wonder if this $182M loss will be reflected in their earnings report on Jan. 22nd.

Tonus said...

Over at Ace's, Paul Demone (who doesn't hide his glee at AMD's misfortunes) linked to this piece that indicates that AMD's cost per wafer is around $4,320. This is compared to Intel's cost of $2,500 per wafer and their smaller die size (at 65nm). As the article states, "[n]o wonder Intel has high gross margins while AMD is losing money."

SPARKS said...

Whew, Tonus, from an above post of mine.

“I told you they were selling garbage”

Obviously, they really ARE trying to sell anything they can off a wafer. According to their financial reports, their net margins are ~ -35%.

Regarding the link and the numbers of layers, copper interconnects inclusive, which they say totals 41 layers, “G”, is there any substance to these numbers?

Moose, you’re right about the AMD/Lehman tank, AMD is down nearly 9% as of 11:00 EDT.

Anonymous, who purchased AMD stock recently, I didn’t pass comment (but I did think it was a ballzy move at the time), I wondered when the Lehman’s issue would surface. The thought crossed my mind at the time, but not enough to give it more thought, let alone a comment. I do know that Barclays did pick the bones of anything of value remaining at Lehman’s a few months back. In perfect 20/20 hindsight, it seems that Lehman’s belly up was more relevant to AMD than anyone expected, especially me. No one wants to buy what seems to be a risk in these times. I don’t fully understand this new “3 Card Monty”, but it can’t be good for the shareholders, but man are these guy’s are crafty.

The only ones who are profiting (?) with this news are the Abduls who will be leaving AMD with less than a 34% stake. That’s cutting the INTC license agreement awfully close (30%) when the deal finalizes.

SPARKS

Anonymous said...

Sparks: Moose, you’re right about the AMD/Lehman tank, AMD is down nearly 9% as of 11:00 EDT.


Down over 11% at close. I'm surprised it didn't tank by double that or more, but there's no understanding AMD investors :).

When exactly does AMD lift the Deneb NDAs? We keep getting some hints about unimpressive reviews, like this one at Hardocp: http://www.hardforum.com/showthread.php?t=1376983. Anyway, I'd bet serious $$ that AMD would love to find a way to NDA their Q4 earnings report! :)

SPARKS said...

“but there's no understanding AMD investors :).”

Yeah there is. Check these numbers.


OppenheimerFunds, Inc. 45,076,288
Maverick Capital, Ltd. 36,125,696

Galleon Management, L.P 32,473,000
AllianceBernstein L.P. 25,303,924

Vanguard Group, Inc. 24,229,368

Credit Suisse Securities (USA) LLC 22,641,130
Barclays Global Investors, N.A. 20,942,712

Halbis Capital Management (UK) Limited 19,443,784

State Street Global Advisors (US) 18,821,226
Capital International Ltd. 10,037,964
Caisse de Depot et Placement du Quebec 7,099,831

Capital Research Global Investors 5,800,000

Citadel Investment Group, L.L.C. 4,934,459
Northern Trust Investments, N.A. 4,723,255
BNY Mellon Wealth Management 4,683,343


http://moneycentral.msn.com/ownership?Symbol=US%3aAMD


The number of shares held by these institutions is 282,335,980. Multiply that by 2.02 dollars a share, we get a little over a half a billion, and these are the big guys. That's 42% of AMD’s total market CAP. Include the little guys and we’re talking a total of 62% of AMD being institutionally held, 608M.

AMD’s market Cap is about 1.23B

Between the cooked up Abdul deal, plus AMD’s held shares, I don’t see any great movement happening which would cause a meaningful price fluctuation. These big guys are sitting tight, waiting on a lawsuit or a miraculous AMD turnaround. They’re in it for the long haul. You could bet your life AMD has nightmares about these big guys getting out. That’s who they’re playing to.

The average daily market volume for a 13 week period was 21M shares. At 2 bucks a share this is peanuts. The tiny guys are causing the small daily price fluctuations, falling in line with this 5 to 10 percent price movement.

This pig is never gonna get off the ground. It’s the institutions that are keeping it on its feet.


SPARKS

Anonymous said...

Here we have proff that Anandtech is intel shill

http://www.amdzone.com/phpbb3/viewtopic.php?f=52&t=135909

Just look at how clicking a big blue Intel advert directs you to information about Intel.

Epic.

Anonymous said...

Twas the night before Christmas

and all was quiet in the fabs as no one wanted them phenom's, tri cores or other garbage

Hector and Dirk await saint blue hoping for High K / metal gate

But alas they have to wait till 32nm at most as the jolly men in blue were too busy balance all their john

Tick Tock Tick Tock AMD is done

Anonymous said...

Ho Ho HO

Merry Christmas

Merry Christmas

Dirk, Hector, Scientia, and Sharikou

What did Santa Bring the sick men in Green?

Ah.. lets peek underneath the Tree

Ohh I see immersion lithography, wow that is really expensive. did it help make transistor go faster? NOPE

Ohh I see the same old boring oxide... damm that is going to result in really leaky gates and poor sub-threshold

Ohh I see low K and lots of metal layers, damm that is expensive and doesn't help too much


Looks like the Grinch stole your Christmas. Maybe next year if you are good them ARABS will loan you another couple 10 billion or so. But heah did anyone tell you oil is now below 40 bucks and they are all running in the red..

Looks like you got the grinch all year around...

Booh Hoo its really one long halloween for ya...

Ho Ho Ho

Ho Ho HO

SPARKS said...

To all, especially the hard working boys at INTC, a very Merry Christmas and a very Happy Holiday. Congratulations on very successful 2008, and hopes for a very profitable 2009 turnaround.

INTC the pinnacle of technological achievement, well done!

Your ever faithful, card carrying, lunatic fringe, balls to the wall, mad dash overclocker,

SPARKS

SPARKS said...

The management at my beloved INTC has given ole SPARKS a cheery little Christmas gift this year, to end the year. I sit here smiling like a Cheshire cat; they have categorically denied NVDA access to anything ATOM. Oh, the joy! Oh, the thrill!

Oh, how terrible, no Nehalem license, no ATOM/ION license. Half the morons are criticizing “monopolistic” INTC for being a mean spirited Scrooge to poor beleaguered NVDA.

Least we forget no license for SLI from 2005 (when it was first implemented) to the end of 2008 on the following chipsets:

x945
955X
975X
x3x
x4x
38X
48X

NVDA on the balls of its ass threw INTC a bone on X58 thinking that it would make everything just dandy. Dead wrong.

http://www.digitimes.com/news/a20081223PD216.html


NVDA went ahead and designed ION anyway expecting a magnanimous and accommodating INTC to cheerfully hand over access to ATOM!!!


The old Sicilian proverb says:

Revenge is a dish best served cold.

NVDA is finding it very cold in the computer industry these days.


SPARKS

Tonus said...

As I understand it, NVIDIA is free to make chipsets for Atom, but Intel is packaging Atom with its chipset and not separately. Any OEM that wants a NVIDIA/Atom solution is welcome to buy an Atom CPU and NVIDIA chipset and swallow the cost of the Intel chipset.

It's a pretty shrewd way to get back at NVIDIA for being stingy with SLI support for so long. I'm sure that the same people who cheered NVIDIA for sticking a thumb in Intel's eye will boo viciously now that Intel has returned the favor. But if you play with fire, you are liable to end up with a few nasty burns. Or perhaps those people can explain to Intel why it should allow NVIDIA to drain profits from them after the way they acted regarding SLI support.

Anonymous said...

Since new year is around the corner here is my predictions for the men in Green

1) Sales trends of Q4 will continue all thru the year as consumers and IT flock to INTEL platform.

2) AMD gets surprised with new yield issues with immersion that IBM didn't see in their low volume fab

3) Arabs cease future investments as the spigot runs dry with oil below 30 bucks

4) AMD cease construction in NY as funding dries up

5) AMD in desperation applies for loans from US goverment but is denied

Boo Hoo poor AMD

SPARKS said...

“Or perhaps those people can explain to Intel why it should allow NVIDIA to drain profits from them after the way they acted regarding SLI support.”

Ah, Tonus, your wisdom surpasses the ancients.

The main reason Nvidia chipsets became an enthusiast object of desire was SLI, of course. Sure, they may have been faster, but there were a plethora of issues that plagued the offerings over the years, stability, quirky drivers, etc. (I am not posting these links as I am sure those who actually worked with these products know exactly what I’m talking about.)

Even if the enthusiast never added a second card, the possibly of adding one in the future was always an option, which is the MAIN reason they sold.

Not so any more, boo hoo, poor NVDA.

Nothing is more stable and reliable than INTC chipsets. Besides, if you want a faster INTC board, just drop a few extra bucks on an ASUS tweaked enthusiast mobo, done. You may have sacrificed some stability (barely) but you still didn’t get SLI. Grrrr!

Personally, that’s what annoyed me terribly for years, NVDA DICTATED my purchase decisions. At every new INTC chipset release from 955X to X48, WEB rumors left me manic with the possibility of SLI support. It never materialized. Time after time, I was left with same compromise on graphics hardware.

Further, I haven’t been alone in this regard.

I’d rather see INTC throw millions at AMD/ATI to aid them in the successful implementation of Crossfire, rather than give NVDA, say, an Arizona weather report (In the desert)

…….And we all know how much I love AMD.

I’ve been waiting YEARS for this payback and I am savoring every delicious moment.


SPARKS

SPARKS said...

"4) AMD cease construction in NY as funding dries up"

From your post to Gods ears.

SPARKS

InTheKnow said...

Finally, we have a system built on the menlow platform that shows some decent battery life.

From the article:
"...the battery life is SO GOOD that you can turn it on with a full battery in the morning, lock the keyboard and screen and it will run for a full work day without it having to go into standby. I’ve done real start-to-finish tests where the i1 has run for over 12 hours. The screen was off and the Wifi was disconnected but with a brush of the mouse pad and a flick of the Wifi switch, it was ready for online action. I’ve also done tests with the device connected to the GSM data network (which takes less power than the higher-speed UMTS networks) and it has shown battery life figures that indicate up to 7hrs connected time. It could be even longer than that as I haven’t done any optimisations to the XP build. The drive doesn’t spin down so there’s obviously room for improvement there and an SSD would probably push it to 8hrs. Maybe there’s even more room for improvements as the device I have here is a factory sample and not a final build.

And this is all with a standard battery in a 500gm handheld package!"


When Moorestown comes out at the end of next year, we should finally see true all day devices and something fit to put into a smart phone.

What does AMD know that I don't? They refuse to play in this space. It still seems to be a natural fit for them since you get so many die per wafer that even with their smaller factory ecosystem they should be able to play on a nearly level playing field with Intel.

Axel said...

Ouch, AMD set to record yet another material goodwill impairment charge for ATI. As before, there will be no impact on cash balance. So of the $5.4 billion acquisition cost, haven't they already written down well over half of it and this will probably make it two-thirds or more?!

Also in Q4 they will record a $70 million hit for downsizing.

Of course, this will probably be just one of thousands of horrific Q4 earnings reports so it might be lost in the noise. If they're going to take a big bath, might as well do it at the same time as everyone else.

AMD took a big bath in Q4 of both 2006 and 2007 as well. At this rate I wonder if they'll still be around to give us an encore in 2009 after Nehalem has effectively commoditized 1P and 2P Opterons?

Tonus said...

AMD also laid off 600 employees, an additional 100 over the amount that they had planned to lay off. That added $20 million to the $50 million restructuring.

The additional writedown on the ATI purchase looks like another way to make the Q4 numbers look a little less worse than they would be otherwise. You start to wonder just when AMD will run out of ways to "cushion the numbers" at the end of each quarter.

Anonymous said...

Tonus: The additional writedown on the ATI purchase looks like another way to make the Q4 numbers look a little less worse than they would be otherwise. You start to wonder just when AMD will run out of ways to "cushion the numbers" at the end of each quarter.

Hmm, I doubt they could write off more than the ATI purchase price of $5.4B, right? Eventually they will have written off the entire investment as "impaired goodwill" or maybe sell it to Goodwill Industries :). I can see the advertising now - GPUs made by the blind, for the blind :)...

Happy New Year everyone!

SPARKS said...

“Eventually they will have written off the entire investment as "impaired goodwill"”

That’s funny. Interestingly enough, the graphics part of AMD is the only thing that’s showing a modicum of success. I’m sure there’s a revenue stream here, somewhere. (We’ll never know)

ATI is relatively successful, yet they still write them down. Looks like another Three Card Monty, again, and again.

SPARKS