9.27.2007

Leaked Toliman Die Image

Lately I have been receiving leaks from inside AMD and Intel from anonymous readers. Some of them I cannot reveal since it could put the person in a lot of trouble due to the sensitivity of the subject matter and their obvious involvement. But some do advise to reveal the secret. Like for instance this leaked image of the soon to be released Tri-Core K10 (code named Toliman). Since this leak was passed on to me by GutterRat from someone else, I cannot guarantee the authenticity of this image.


38 comments:

Andy said...

lol

What's funny is that with the way AMD have been operating lately that could easily be their image for the press and such.

Anonymous said...

*lol*

Anonymous said...

Which one is the disabled core?

Anonymous said...

*rriiiinnng*... *a girly voice whispers*, "seven days".

Anonymous said...

Scientia says that this is a good thing, it means the Tolima will use up less die space.

Anonymous said...

OK - I'm never posting on this blog again - there was a promise not to reveal this trade secret information and you blew it! Seriously, though - pretty damn funny!

My only question is what is that "feature" on the fourth core? Or is that "AMD innovation"? Is that FUSION? It is isn't it!?! Now I understand where fusion comes from!


Judging from the size of the shadow, the shape of the defect, the fact that it is the last Thursday in September and my vastly superior knowledge to you mere mortals, I would judge that dies size to be 283mm2 give or take 1mm2! Can someone please help me plug #'s into the die model so I get this result!?!

I'm waiting for Abinstein to claim the die size is now smaller because you can exclude the defective area of the fourth core from the die area calculation! (You know because it is not a functional part of the chip, just like the scribe line, it doesn't count!)

And for those, who hadn't heard - if all 4 cores are working and AMD needs more tri-cores they will simply cut out one of the functional cores, sell a tri-core and sell the remainder as well! Codename is "unicorn", or was it "uni-core", I keep forgetting... however the full name will be unicorn "black edition"...

Andy said...


My only question is what is that "feature" on the fourth core? Or is that "AMD innovation"? Is that FUSION? It is isn't it!?! Now I understand where fusion comes from!


It is a "Black hole" that sucks all the computing power in the world into a 283mm2 chip. FASN8ING shit

SPARKS said...

Wait a minute, can that ‘defect’ be for real? Has anyone out there ever seen anything, remotely, like this before, for real? If so, what would cause such a travesty?

God all mighty, is that ugly.

SPARKS

Anonymous said...

XD

Anonymous said...

This isn't "native". DAMN AMD!!!

GutterRat said...

roborat,

Aww, shucks :)

Anonymous said...

It's an exposed processor keychain. Except theres no loop... Or chain for that matter.

GutterRat said...

If you look close enough, this makes perfect sense.

The hole can now be filled by the green ring/donut/blood flow restraint that's pervasive in the new Opteron logo.

Anonymous said...

Correction

It's "Toliman" not "Tolima"

Rhymes with "Taliban"

Anonymous said...

Everyone,

Looks like they boys at AMDZone have put up a poll to ask whether Hector should stay or go.

The numbers are pretty close.

http://www.amdzone.com/index.php?name=PNphpBB2&file=viewtopic&t=11280&start=0&postdays=0&postorder=asc&highlight=&sid=1c5cad5883950cd5cd953337ad573290

Anonymous said...

Interesting blog entry from Anand, mostly positive on Intel

Link

What caught my eye

I talked to AMD about triple core, which will be coming in the beginning of 2008 as quad-cores with one core disabled. It doesn't look like triple core will ever be its own design, so the product will last as long as it can. Either quad-core will get cheap enough to make triple core obsolete or triple core demand will be high enough to warrant it sticking around, but I figure it'll end up going away. Until then it should offer an interesting affordable alternative to dual-core, unfortunately we'll have to wait until next year to see it.

If above is correct, then the leaked "taliban" die image is right on, despite speculation about a ground-up "tripod" design from AMD.

Anonymous said...

"...despite speculation about a ground-up "tripod" design from AMD."

The speculation is from idiots who have no background - AMD would have to completely re-layout the chip go thru several tapeouts, development Si, additional engineering costs and then have to maintain an additional product in the fab which leads to operational efficiencies (switching masks between product lots, etc..)

I guess there were some idiots that though AMD could just make it triangular-ish? Of course when you have to cut die (and print them) you really need rectangles or squares...

Anyone who though this would be a ground up design for more than 10 seconds should be put into the "auto-disregard, the guy's an idiot" category.

pointer said...

Anonymous said...

"...despite speculation about a ground-up "tripod" design from AMD."

The speculation is from idiots who have no background - AMD would have to completely re-layout the chip go thru several tapeouts, development Si, additional engineering costs and then have to maintain an additional product in the fab which leads to operational efficiencies (switching masks between product lots, etc..)


No, that is not true. AMD's tricore is actually a ground up design, and we have already seen it .. it was called Barcelona! AMD designed it in such that there is an extra core ready to be fused out when thing gone wrong on one core so that they would have extreme good yield ... and with any luck, they would have some supply of quad core too ...

[end BS]

GutterRat said...
This comment has been removed by the author.
GutterRat said...

pointer said,

No, that is not true. AMD's tricore is actually a ground up design, and we have already seen it .. it was called Barcelona! AMD designed it in such that there is an extra core ready to be fused out when thing gone wrong on one core so that they would have extreme good yield ... and with any luck, they would have some supply of quad core too ...

Here are the facts:

1) It is the poor yields of 65nm rev 10h cores that have motivated the "Tripod Taliban" salvage operation. Har.

2) Your assertion that the "Tripod Taliban" is a ground-up design is laughable. Har har.

3) The notion that news of the alleged "Tripod Taliban" was kept a closely guarded secret from partners for "competitive reasons" is disingenuous. More than a few inside canaries are singing BS on this one. Har har har.

4) Which customer demanded production of the alleged "Tripod Taliban"? DAAMIT! Thou shall not engage in non customer centric innovation for that is your motto.

pointer said...

GutterRat said...

pointer said, ...

....


errr, i was BS-ing ... as indicated by the [end BS] ... I tot i can do this in the humor post ... :)

GutterRat said...

pointer,

I saw the [end BS] at the end of the message but it was worded in such a way that made me conclude differently.

No harm, no foul...I hope :)

pointer said...

GutterRat said...

pointer,

I saw the [end BS] at the end of the message but it was worded in such a way that made me conclude differently.

No harm, no foul...I hope :)


that's ok. I'm professional, even when BS-ing and your response just proved how professional I am. :)

Anonymous said...

It's just that no one expects BS in a discussion about AMD.

Roborat, Ph.D said...

It's "Toliman" not "Tolima"

updated, thanks.

It's just that no one expects BS in a discussion about AMD.

true. we're very serious here.

Giant said...

AMD estimates cut yet again:

http://www.eetimes.com/news/semi/showArticle.jhtml?articleID=202102862



''We believe the company's late Barcelona introduction and disappointing early performance are an early indication of a bad marriage of process technology and design that will be hard to fix before a move to 45-nm is required,'' said analyst Doug Freedman of American Technology Research Inc., in a new report.

''AMD still has a lot of work to do to fix the architectural mismatch of Barcelona with the 65-nm process node and the poor performance of R600,'' he said, referring to ATI's latest graphics chip. That chip is also late to the market.

In response, AMD is expected to move the attention over to a 45-nm version of the quad-core processor. ''We expect a shift in attention to AMD's 45-nm pilot results and demonstrations as early as December at the company's analyst event, where we believe it will be more important to deliver early 45-nm samples than a 2.5GHz Barcelona processor,'' he said.

''There is no doubt that new process and product designs increase the risk of success, but in this case it appears to be the only option and best way to regain competitive products,'' he said. However, ''Intel's execution/roadmap puts even more pressure on AMD to deliver 45-nm on time in summer 2008.''

As a result of the problems, Freedman lowered his estimates on AMD. ''We are lowering our Q3 revenue and EPS estimate to $1.529 billion and a loss of $0.62 from $1.557 billion and a loss of $0.58,,'' he said. ''For 2008, we now forecast revenue of $6.855 billion verses our prior expectation of $7.237 billion, with our net loss estimate lowered to $0.97 from $0.44.''


AMD is finished. Running of cash rapidly, and losses are expected to continue well into 2008 until BK.

Anonymous said...

''We are lowering our Q3 revenue and EPS estimate to $1.529 billion and a loss of $0.62 from $1.557 billion and a loss of $0.58,,''

Sharikou was overheard saying that "this is great news! AMD is only going to lose sixty-two cents for the quarter! The turnaround is complete! Intel BK tomorrow!!!"

Anonymous said...

Is Fudzilla reliable?

Yorkfield 3.2GHz to launch at FSB 1600

Fudzilla

Poor AMD fanbois if this turns out to be legit.

Anonymous said...

Honestly, the cut in estimates is a bit surprising - indications are that this should be a strong quarter overall for the x86 market (and Q3 is also in general seasonally better than Q2).

Putting aside the crap about Barcelona (it won't potentially move AMD's bottom line until Q1'08), they should have sold more processors and have better (lower) losses...unless ASP's got whacked again?

I'm also surprised about the net loss estimates for 2008! Wasn't AMD forecasting getting even by Q4'08? This gets kind of scary as money is no longer freely available and with no real profit in sight, AMD's cash flow will remain negative.

Perhaps AMD needs to bite the bullet and stop slashing prices to gain market share? (Or just get rid of Ruiz and put in someone who knows how to run a company)

I guess we'll get some insight in to how good or bad things are with earnings reports coming up - Intel (Oct18) and AMD (Oct 20?)

Andy said...


Is Fudzilla reliable?

Yorkfield 3.2GHz to launch at FSB 1600

Fudzilla

Poor AMD fanbois if this turns out to be legit.


What is funny is how Fuad (or FUD as we named him) has changed his tune totally.
He used to be very close to the AMD side. Now he seems to embrace Intel without question. I think he has looked at the situation and realised only a complete idiot would say AMD are better.

SPARKS said...

Is Fudzilla reliable?
Yorkfield 3.2GHz to launch at FSB 1600




C2QXEE, X48 @1600, hmmmm.

Anyone wanna guess what’s gonna be under MY Christmas tree???

Hmmmm, delicious.

SPARKS

Lem said...

All of you laughing at AMD for this tricore, what would you do with a quad core die that had 3 good cores but one not-so-good core? (eg. 3 cores might clock to 2.5Ghz, but the 4th only clocks to 1.5Ghz reliably .. or, 3 cores are good, but one has a fatal flaw)

Would you (from the example above. 3x2.5Ghz, 1x1.5Ghz):

1. Sell it as a 1.5Ghz quad-core
2. Sell it as a 2.5Ghz tri-core.
3. Sell it as a 2.5Ghz dual-core.
4. Sell it as a 2.5Ghz single-core.
5. Throw it in the rubbish bin.
6. Take it home and eat it. Mmm, chips :)

Seriously though, from options 1-5, which would you do if you were AMD with quad-core K10, or Intel with quad-core Nehalem?

We can talk about native octo-cores (Nehalem too?) later ;)

The truth is, yields aren't perfect, and with a die like K10 on 65nm, or Nehalem on 45nm, there's bound to be 3 good cores and one not-so-good core. Intel might be able to get away with selling all of it's dodgey quads as duals due to its sheer size, but AMD needs to make every dollar it can, and tri-core is a good way to do it, IMHO. If you disagree, say so and why.

SPARKS said...

Well, apparently you’re new to this site, or you haven’t been following it very closely. Additionally, even if you were following the site closely, you missed Intel’s response to the same scenario(s) you posed was answered a few months back. Intel’s response was,’ it would very difficult if not impossible to do native quad now.’ Further, Paul Ottelini response speaks volumes, “we like to sell processors which have all the cores working.” Interpreted as an arrogant response, it the message is clear, I am sure HE would choose number (5) of your hypothetical product/development, cover thy ass, options.

Then again, Intel was smart enough to avoid these options from the onset, which renders the supposition academic. AMD should have done the same. The worst of it is, they thought mistakenly, I might add, they could fix these issues, while in semi production. This was a fatal flaw in the plan. Now AMD is faced with all the options you itemized.

Hartcourt Fenton Mudd, “death by hanging, death by phaser, the key word Mr. Spock is DEATH!”

Therefore, number seven should be ‘don’t get involved with a new architecture, a new process, and huge die in the first place.

This is perfect 20-20 hindsight, proven by AMD’s top exec’s bailing; you can’t go to warp speed with a hole in the core. They just beamed out to another ship.

It is rather presumptuous to assume ALL of us are laughing at AMD’s monumental blunders, Dead Wrong! As a Union Worker (IBEW LU#3, NYC), I absolutely despise hard working, talented people losing their jobs due upper managements WRECKLESS decisions.

If they give me a plan that doesn’t work, I’ll throw it back at them, and tell ‘em this ain’t gonna work, go sharpen your pencil! No one listened to the poor bastards at AMD who probably knew that this pig wasn’t gonna fly. They did it anyway, and rammed it right down their throats.

This is not funny, pal, it is almost criminal. Go ask the AMD shareholders and employees what they think.

So much for your frivolous post.

SPARKS

Anonymous said...

All of you laughing at AMD for this tricore

I think most people are needling AMD for making it seem as if tri-core is in response to customer demand. This is a natural response for a business, since admitting that a tri-core is really a quad-core with a disabled core would be bad PR. But they were bound to be chided about it.

Although yeah, AMD's supporters have poked fun at Intel for it's non-native approach to quad-cores, but Intel seems to have taken the practical approach, and it appears to be paying off. If AMD is indeed having yield issues due to trying for native quad-core from the start, Intel will have taken the smart approach,and AMD will remain the butt of jokes.

But as for your question, if I was in their situation I'd do the same thing they're doing: sell the "bad" quad-cores as tri-cores and try to paint it as a good thing, even if people can see through the BS.

Anonymous said...

"ll of you laughing at AMD for this tricore, what would you do with a quad core die that had 3 good cores but one not-so-good core?"

Assuming "yields are mature" and "yields are good" this should be a small percentage of parts, no? (If AMD is telling the truth that there are no production issues with Barcelona and it is only design issue).

So, if this is true I would sell the part in your scenario as a DUAL CORE! Yes, it sounds crazy, you could possibly get more from a "tri-core" but there are a couple of advantages to this approach:

1) Fewer SKU's to maintain, these could be mixed in with the dual core bin (with a different revision # of course) - this means a couple of less parts to worry about stocking levels and easier to predict production levels.

2) It eliminates the public perception that your latest design is "defective" - tri-core screams quad core with one core not working or performing badly! It doesn't scream innonvation or competitive advantage over Intel! A downbinned dual core would get lost in the shuffle of "native dual core parts" and avoid the PR nightmare.

3) It eliminates the potential issue if there is actual demand for the part? What if there is more demand then you have defective parts? Do you start downbinning GOOD quad cores? Design a tri core from scratch (absurd)? Just tell customer to stick it with their tri-core demand and focus on higher revenue parts?

Let's face it, this is likely a temporary part - and it only makes sense if you HAVE A LOT OF DEFECTIVE QUADS! If this is a small and temporary issue (ahem, as AMD wants every one to believe) then you are better off binning them as duals and avoiding the headaches mentioned above.

You need to look at it from a business perpective and tradoff the cost of maintaining MULTIPLE new SKU's (or are you just going to only have one clock speed for tri-core?) vs the volume of product and then factor in the strategic impacts mentioned above. If we are talking >1 million parts, then maybe...

One last thing to consider - AMD has indicated they are ONLY DOING THIS INITIALLY ON DESKTOP! Ask yourself how much volume is quad core on desktop to begin with and the from that figure out how much volume of that will get downbinned into tri-core... is this a mainstream audience? Enthusiast audience (who you'll want to sell quads to anyway)?

Anonymous said...

LEM - One additional option:

7) Only manufacture dual cores in volume until you can figure out the quad core problems!

AMD - has continuously fed the BS that they only launch/ramp products when yields are mature? Well quad parts are obviously not mature or yielding well!

REMEMBER this is desktop only for tri-cores... AMD should just manufacture dual cores until they figure out how to get quads working well in volume. If this means the need for a smaller die, 45nm is only 6-8 months after quad phenom launch! (if you believe the 45nm BS AMD is shovelling too!)

Finally look at the PROFIT MARGINS on dual core vs tri core parts! Yes tri-core will sell for more than dual core parts, but it is double the Si area and therefore nearly double the production cost (not quite as packaging cost will not be double).

So are you going to be able to sell tr-core for double the price? Or are you better off making more dual cores and earning a better margin on them. Your last point indicates AMD needs every possible dollar - well on desktop their margins will be best on dual cores until the get quads working at higher yields...

Anonymous said...

Let's put some imaginary numbers to the comment above:

Two 2.4GHz dual core parts:
cost = $75(each)
price around $150each
Profit = $150 total (for 2 die)

One Quad core part 2.4GHz:
Cost = ~$140 (not quite double), price around $300
profit = $160

Now lets factor in 30% tri-cores due to bad yield (and we'll take the overly optimistic assumption that there is no additional yield loss, the only impact is downbinning to tri-cores - this obviously is a RIDICULOUS assumption)

Let's say tri core is 1/2 between dual and quad - price ~$225, cost remains $140 so profit is now $85.

So 70%*$160 + 30%*$85 =
~$137 profit vs $150 profit for selling 2 dual cores!

Now consider overall yield of quad will be reduced (not just downbinning to quad), only 30% may be optimistic? (if it is lower than 30% why even bother with tri-core, might as well take best 2 dual cores and bin a a higher dual core part)

LEM: The whole problem with your speculation is you are trying to put a simple and straightforward analysis ("might as well try to get every dollar you can") to a complex problem:

1. There are technical issues we don't understand; would binning quad all the way down to dual core bump up the clockspeed bins and therefore there might not be as much of a price drop off between that and a lower clocked tri-core part.
2) what is the cost of maintaining these stock (there are inventory, planning, logistics, marketing and all sorts of hidden costs noone considers)
3) Is this a long term plan? Or is this only until yields get better? (is the channel going to get shafted with a bunch of tricore inventory when all is said and done?)
4) what are the incremental cost of fusing (probably marginal) and re-testing/re-bining/qualifying the part as a tricore (not trivial?)
5) what is the market impact to th epricing levels of other parts? Tri-core puts a floor on the quad (hard to sell a quad for lower price than a tri unless it is substantially lower clock speed). Does it also put a CEILING on dual core parts? How much higher does a dual core part have to be over a tri-core to justify a higher price.

#5 is a difficult marketing problem - Intel already has this issue with the overlap in pricing between dual and quads - you are now introducing 2 more pricing overlaps scenarios (dual to tri and tri-quad)

So my guess remains this is a short (AMD hopes)term deal and AMD wants to milk more money out of their desktop fanboy base. Unless quad yields are truly horrendous (which may be the case?), AMD is better off selling them as dual cores.

Anonymous said...

I thought AMD's Triclops would have a big, steaming turd from Hector sitting on one of the cores