[SemiAccurate] Nvidia's Fermi GTX480 is broken and unfixable

[bryanW1995]

maybe he can claim that intel told him that larrabee was doa...

 

[cbn]

Deleted

 

[cbn]

Deleted (double post)

 

[C@rnage]

hey guys im new here.so i read this article about nvidia's new fermi architecture falling on its face.some ree-ree sitting in the back of a small yellow bus wrote it,his name was charlie.look

guys,this article is crap,everything charlie writes/says...is crap.obviously weve all seen the demos of fermi spanking some ass @ CES 2010.anyone anticipating this launch should not be scared

away by the likes of charlie.i watched the farcry 2 demo ---> here http://www.youtube.com/watch?v=yNbY-m28bUk where "Fermi" the gtx 480 got 84ish fps with everything maxed out on a 24"

resolution.i watched the HD5970 benchmark ---> here http://www.youtube.com/watch?v=K8gjrgMOEqQ almost the same exact specs except fermi only had 4x AA and the 5970 had 8x.big

whoopdy wow wow.the 5970 got 80 ish fps.if you pay any kind of attention youll see the fermi had higher minimum fps. fermi had 64fps min and 5970 had 52 fps min.cmon guys,not to use cliches

but....the proof is in da puddin.yes,im an nvidia fanboy.i love what ATi is putting on the market right now A+ to them.to each their own.before anyone posts something stupid like im a "Nub" or

that "I dont know anything about computer hardware"...i built a 20,000 3dmark06 computer with a pair of 8800GT's and an E8500.thanks.



this isnt my best score just some proof http://service.futuremark.com/resultComparison.action?compareResultId=11226543&compareResultType=14

 

[shangshang]

^ C@arnage is a cousin of Wreckage? anyway I hate fanboy lol

 

[Schmide]

that "I dont know anything about computer hardware"...i built a 20,000 3dmark06 computer with a pair of 8800GT's and an E8500.thanks.

Well I built a reputation with tawdry women and cheep wine.

Discuss the article and not your epeen.

 

[jvroig]

I enjoy the forum posts at semi-accurate more than the articles. The articles are (perhaps understandably) sensational, but in the forum, I have read quite a number of posts from Charlie where he does in fact (rather surprisingly?) sound very very sane and logical. (One has to wonder, perhaps Charlie doesn't "want" to bash nVidia, but seeing as to how it is the lot given him, and he has to make a living, perhaps just goes about it with passion, with panache... or of course, he could really just be an nV-bashing-jerk that he appears to be from his articles. I don't know him personally to make a clear judgement)


While I am not swayed by Charlie's article, I have to admit that Anand's article about the "inside story" of RV870 (oops, sorry, I meant "Cypress") had me suddenly worrying about Fermi. Maybe it is purely the fault of TSMC, or maybe they share it with nVidia, right now for us consumers it is not really as important as the fact that there does in fact seem to be a problem (they would not be this late to launch willingly), and perhaps the only thing that is really allowing nVidia to save face in all these is that they have not announced any official launch dates (as Keysplayr has pointed out in more than one occasion), so all delays so far are merely "unofficial delays" I suppose. Still, not exactly good for us. And seeing how ATI's cards are getting gobbled up, not exactly good for nVidia either.

^ C@arnage is a cousin of Wreckage?

Wreckage is far more coherent, and each Wreckage post is properly formatted (to the best of my knowledge). Relating this one to Wreckage is an insult to all Wreckages around the world.

.the proof is in da puddin

It's "the proof of the pudding is in the eating". When you say it as you did, it makes absolutely no sense.

guys,this article is crap,everything charlie writes/says...is crap

Thank you, do you mean that Fermi is not delayed? Or that Fermi is not too hot? Or that Fermi is not suffering from delays due to being a larger and probably more complex card than it's ATI counterpart? Charlie is hardly the best source for "fairness", that much is transparently obvious, but when you say "everything", then please do enlighten some of us who wish to know which ones of those are wrong and what is the truth that Charlie forgot to mention / covered-up.

anyone anticipating this launch should not be scared

Then please give us a reason. The article from Anand (which is hypothesized to be the "sole" source of this Charlie article) does in fact point out to big problems with Fermi, for both TSMC and nVidia's fault (who is to blame more is debatable, except for when nVidia, probably in desperation, called for 'zero via defects', which is just easy pickings for a good laugh). And that Fermi has been delayed (unwillingly, of course; I can't imagine nVidia would actually plan for Fermi to miss the Win7 "bulge" and then also continue to let ATI's new cards lord over the market without any answer) also does not paint a very rosy picture. You seem to know the answer though, so please, why is it that we should have no fear? Is it because Fermi's bugs have finally all been worked out in A3 silicon? Are there in fact no heat issues? Is defect density a problem of the past for Fermi already despite the huge die size? Or more importantly, are you confident launch price will be reasonable and product quantity at launch will be enough?

 

[solofly]

^ C@arnage is a cousin of Wreckage? anyway I hate fanboy lol

lol

 

[Dark4ng3l]

C@arnage's post is so poorly formated that it has to be a botched copy/paste. If he somehow did that while actually typing his message in the box then I have no words to describe how epic his failure is....

 

[Soleron]

When nvidia finally gets fermi out the door in numbers, Charlie will have to find a new target.

Can I have a go?

Fermi II is late, hot, unmanufacturable, and oh did we mention unmanufacturable? And late, and hot. My sources tell me each Fermi II will cost Nvidia $5000 as via defects, stray elephants and Dear Leader personally jumping up and down on each wafer will mean only one die will work per wafer. So let's add a reasonable upper bound of $2000 for the board (And I'll assume AMD gets all their memory for free, so $0) and $996 for wood screws in every Fermi II board, and we arrive at a (ballpark estimate, I've simplified the math) of $1m per Fermi II. Meanwhile AMD's total cost is still $0, trapping Fermi II's price in a 'box' between $0 and paying people $50 to take them away.

Since this is the C7946 stepping we can add four weeks for tapeout, seventeen weeks for tea breaks, nine weeks for pulling out the dead transistors with tweezers and another week for generic unspecified incompetence on the part of Nvidia. So that puts boards into next century for a paper launch and they're only producing one batch so after reviews there will be less than zero cards on shelves. Factor in how late and hot and late it is and you can see that Nvidia's HQ is going to crash into the sun in a few years.

 

[GaiaHunter]

Can I have a go?

Fermi II is late, hot, unmanufacturable, and oh did we mention unmanufacturable? And late, and hot. My sources tell me each Fermi II will cost Nvidia $5000 as via defects, stray elephants and Dear Leader personally jumping up and down on each wafer will mean only one die will work per wafer. So let's add a reasonable upper bound of $2000 for the board (And I'll assume AMD gets all their memory for free, so $0) and $996 for wood screws in every Fermi II board, and we arrive at a (ballpark estimate, I've simplified the math) of $1m per Fermi II. Meanwhile AMD's total cost is still $0, trapping Fermi II's price in a 'box' between $0 and paying people $50 to take them away.

Since this is the C7946 stepping we can add four weeks for tapeout, seventeen weeks for tea breaks, nine weeks for pulling out the dead transistors with tweezers and another week for generic unspecified incompetence on the part of Nvidia. So that puts boards into next century for a paper launch and they're only producing one batch so after reviews there will be less than zero cards on shelves. Factor in how late and hot and late it is and you can see that Nvidia's HQ is going to crash into the sun in a few years.

Isn't that a joke from about September 2009 when Charlie was saying that Fermi would only be out Q1 2010 best scenario and most likely Q2 2010 and people were saying "Meh it's only Charlie!" and of course Fermi will be out before the end of 2009?

Guess the world is full of "Charlies".

 

[Keysplayr]

Charlie has run through many scenarios over the past 8 months. They include various problems that could arise, different times for launch depending on if this happens or that happens. Many if/thens. When the time does come and GF100 finally does hit the shelves or at least reviewers hands, Charlie will simply go back to one of the 17 scenarios he went through over the last 8 months and claim he was on or near his mark. Simple. Eventuality would dictate that one of his claims would be closer than another. And he'll pick the closest. Natch.
This is all he does folks. And it's eaten up, unfortunately.

 

[GaiaHunter]

Charlie has run through many scenarios over the past 8 months. They include various problems that could arise, different times for launch depending on if this happens or that happens. Many if/thens. When the time does come and GF100 finally does hit the shelves or at least reviewers hands, Charlie will simply go back to one of the 17 scenarios he went through over the last 8 months and claim he was on or near his mark. Simple. Eventuality would dictate that one of his claims would be closer than another. And he'll pick the closest. Natch.
This is all he does folks. And it's eaten up, unfortunately.

Its true.

It doesn't change the fact Fermi is delayed compared to early expectations, though.

Considering it is a fact that Fermi is late, unless you prefer to believe NVIDIA was planning on only launching Fermi somewhere in the near future, Charlie' scenario makes sense to explain the fact until its scenario is proven wrong or some other scenario is proven right.

As you say, many IFs/Then. But that is ok. Although he prefers to focus on the ones that are darkest for NVIDIA - dark scenarios sell more after all.

 

[jvroig]

Considering it is a fact that Fermi is late, unless you prefer to believe NVIDIA was planning on only launching Fermi somewhere in the near future, Charlie' scenario makes sense to explain the fact until its scenario is proven wrong or some other scenario is proven right.

Agreed. Charlie or no Charlie, it is very easy to believe that there were major problems (and Anand's article on the RV870 back-story hints at a few probably), I have no doubt nVidia never wished to launch this late in the game.

 

[Lonyo]

Charlie has run through many scenarios over the past 8 months. They include various problems that could arise, different times for launch depending on if this happens or that happens. Many if/thens. When the time does come and GF100 finally does hit the shelves or at least reviewers hands, Charlie will simply go back to one of the 17 scenarios he went through over the last 8 months and claim he was on or near his mark. Simple. Eventuality would dictate that one of his claims would be closer than another. And he'll pick the closest. Natch.
This is all he does folks. And it's eaten up, unfortunately.

Equally he is also tells us things well in advance of places like AT.
http://www.semiaccurate.com/2009/06/20/why-are-there-no-4770s/

There have been a lot of claims about low yields for RV740, some as low as 20%. Having heard much more exact figures, lets just say those numbers aren't even close, yields are much higher than that. ATI ran a full production run of wafers, and got tens of thousands of working parts back. The reasoning behind running the wafers was simple, TSMC 40nm has many problems, some of which aren't seen until you are in a volume run, was to figure all out the problems out.

So, ATI ran RV740 knowing that yields would be low, and that they would have problems. If the problems were solved during the run, more could be put in the oven. If not, the Plan B was to use an R770 variant fused off to fill that market segment. There wasn't a real down side to it other than perception.

The result of the RV740 test is that TSMC 40nm did indeed have problems, and yields were indeed low. The places that problems lurked were illuminated, and that knowledge was rolled into the whole Evergreen family. Given the lead times between the two families of parts, and the long gestation periods, we are told it gave Evergreen quite a nice leg up. If you look at the simplified version of yield, basically a fixed distribution of defects on a wafer, a 50% yield on a 135mm^2 chip means almost zero on a chip double the size, and essentially zero on something 3x as large. Reality doesn't mirror that, but it is a good starting point for arguments.

This production run was essential for ATI, and it makes you wonder about Nvidia and how they are going to cope with the same problems given the curious disappearance of the GT215. If they are doing the same thing with the 216 and 218, that is smart, but a 500mm^2 part is still going to be a long stretch.

This is essentially what Anand recently wrote in his ATI article, only Charlie wrote his version in June of last year.

Maybe someone could do an analysis of Charlie's Fermi articles and see what his expected release dates were from the start. I think there was one from May of last year which claimed a March or so launch date for Fermi, so he was fairly on the ball >9 months ago with that articles estimate, but it depends on how much he hedged his bets.

 

[Soleron]

Charlie has run through many scenarios over the past 8 months.

Really? Can you quote some of his other scenarios that haven't panned out? All he's done is said best case/average case/worst case with a justification for each and what Nvidia needs to do to hit it. Last May his estimate was March, and he's been consistent about late Q1 for the average case right through to this article.

It's not nearly as bad as Fudzilla who really have said 8 different release dates, without any sort of reasoning or explanation for changing it.

 

[Kuzi]

For the most part, it seems Charlie was correct about TSMC's 40nm issues and the Fermi delay. He did say many months ago that Fermi would not be released in 2009, and from what I can recall, he did mention March as the probable release date for Fermi.

 

[SlowSpyder]

Really? Can you quote some of his other scenarios that haven't panned out? All he's done is said best case/average case/worst case with a justification for each and what Nvidia needs to do to hit it. Last May his estimate was March, and he's been consistent about late Q1 for the average case right through to this article.

It's not nearly as bad as Fudzilla who really have said 8 different release dates, without any sort of reasoning or explanation for changing it.

Yea, I thought Charlie was pretty consistent with his articles about Fermi. And correct it seems, at least regarding it's release. I think Fudzilla is the one who posts every possible scenario.

 

[Soleron]

I've just gone through every big article since May, and his claims are pretty consistently March for the average case. No contradictions I can see.

A look at the Nvidia GT300 architecture (May 14, 2009)
Miracles happen, GT300 tapes out! (July 29, 2009)
Four more GT300 variants tip up (August 5, 2009)
GT300 to have an NVIO chip (August 13, 2009)
Nvidia GT300 yields are under 2% (September 15, 2009)
Nvidia kills GTX285, GTX275, GTX260, abandons the mid and high end market (October 6, 2009)
Nvidia finally gets Fermi A2 taped out (November 2, 2009)
Fermi massively misses clock targets (November 16, 2009)
Fermi A3 silicon is in the oven (December 10, 2009)
Nvidia castrates Fermi to 448SPs (December 21, 2009)
Nvidia GF100 pulls 280W and is unmanufacturable (January 17, 2010)
Nvidia's Fermi GTX480 is broken and unfixable (February 17, 2010)

--

A look at the Nvidia GT300 architecture (May 14, 2009)

- Nvidia says October 15 launch date, he disagrees
- With no spins and perfect execution, he says November
- With one spin, you're into next year (into Q1)
- With two spins, you'll have a hard time hitting Q1
- Two spins is the most likely looking at past products
- Won't come close to R870 [This means the dual-card which we now know as 5970, not the RV870 5870]
- Whole article focuses on how the chip is focused on GPGPU, many months before Nvidia announce it as such

Miracles happen, GT300 tapes out! (July 29, 2009)

- Late July tapeout
- Mid-December for first parts on shelves if no new spins, add 8 weeks per spin [1 spin means February, 2 spins

means April]
- Die size of 530mm^2
- "Vapourware for 2009"

Four more GT300 variants tip up (August 5, 2009)

- Four derivative parts, none taped out by then

GT300 to have an NVIO chip (August 13, 2009)

- GT300 has an NVIO chip

Nvidia GT300 yields are under 2% (September 15, 2009)

- First A1 silicon back in early September
- Low single-digit yields for A1 stepping
- 104 die candidates per wafer

Nvidia kills GTX285, GTX275, GTX260, abandons the mid and high end market (October 6, 2009)

- All mentioned parts soon to be EOL [in retrospect, he's still saying they did stop production of new GT200 parts and what we see in the channel is very small numbers of warehouse stock coming out, which matches up with the poor stock situation]
- No Fermi derivatives taped out
- Now claiming 530mm^2 to be a minimum estimate, now claims 23.x by 23.x mm^2

Nvidia finally gets Fermi A2 taped out (November 2, 2009)

- A2 taped out a few weeks previously; A2 samples in December
- Risk wafers. If A2 was OK then February [matches with earlier estimate for one spin]
- If A3 needed then put into March for samples/paper launch
- Claims Nvidia is estimating March internally, implying he believes March to be the most likely date
- No Fermi derivatives taped out

Fermi massively misses clock targets (November 16, 2009)

- Still claiming >530mm^2 die size
- 20% clockspeed miss [taking into account later claims of 1500MHz shaders being the internal target, that means 1200MHz.]

Fermi A3 silicon is in the oven (December 10, 2009)

- Nvidia begins an A3 stepping
- First silicon to be back in early January
- If risk wafers are still valid, then late February (6+2 weeks). If not, then late March (10+2 weeks).
- First A3 silicon able to be shown at CES (no guarantee, just estimating from timing)
- Best case: Feb 1; Worst case: April 1 Average case: mid-March. Thinks A4 is unlikely.
- Clocks for A2 are 500MHz half-core implying 1000MHz, but A3 will up clocks.
- Says that a 500MHz/1000MHz/512SP Fermi would barely beat the 5870
- 'Alarming' yields on A2

Nvidia castrates Fermi to 448SPs (December 21, 2009)

- Fermi will only have 448 SPs on Tesla variants.
- Says that a 448 SP version will only barely fit in 225W TDP
- Speculates that since Tesla is really expensive and small volume, you'd think they would have their best bins in there. As a result, he doesn't expect any consumer versions to also be only 448 shaders and hot.
- Fermi may beat Cypress by a little, but more expensive and much hotter

Nvidia GF100 pulls 280W and is unmanufacturable (January 17, 2010)

- Revised die size estimate of 550mm^2 (on forums he says initial estimate was only 23.x mm^2 by 23.x mm^2, not precise, and he said the lower figure earlier but new sources mean that .x is higher so therefore ~550.)
- Repeats claim of 104 die candidates per wafer
- Says a 512SP version of GF100 would pull 280W.
- Says initial GF100 chips will be 448 shaders and downclocked considerably. Initial target was 1500-1600MHz but shipping cards will be 1400MHz halo part and 1200MHz 'volume bin'.
- Mentions the 'voltage versus amperage' thing contributing to high power consumption
- Says no Fermi derivatives will come out before ATI completes its Evergreen lineup
- AIBs will get samples in late Feb, March launch

Nvidia's Fermi GTX480 is broken and unfixable (February 17, 2010)

- Top bin will launch with 448 shaders and 1200MHz shader clock. A3 did not up clock speeds.
- Single digit yields
- Internal expectations were 1500-1600MHz.
- Reaffirms 280W power draw.
- Estimates performance of 448 shaders / 1200MHz as 12% better than the 5870

 

[GodisanAtheist]

Equally he is also tells us things well in advance of places like AT.
http://www.semiaccurate.com/2009/06/20/why-are-there-no-4770s/

Holy crap a charlie article where I could imagine the author without froth around the corners of his mouth! Charlie should try that writing style (maybe he took some diazepam?) more often, and his articles would begin to have that sheen of credibility his vitriolic ranting lacks.

 

[Soleron]

Holy crap a charlie article where I could imagine the author without froth around the corners of his mouth! Charlie should try that writing style (maybe he took some diazepam?) more often, and his articles would begin to have that sheen of credibility his vitriolic ranting lacks.

Read his forum posts. They say almost everything the articles do but in the improved tone.

 

[tviceman]

Miracles happen, GT300 tapes out! (July 29, 2009)
Nvidia kills GTX285, GTX275, GTX260, abandons the mid and high end market (October 6, 2009)
Nvidia castrates Fermi to 448SPs (December 21, 2009)
Nvidia GF100 pulls 280W and is unmanufacturable (January 17, 2010)
Nvidia's Fermi GTX480 is broken and unfixable (February 17, 2010)

Those are very, very, sensationalist, misleading, and biased. If you don't like someone/something, you will always find something to slant or bad to say whether it's true, false, fact, opinion, or speculation.

Nvidia GT300 yields are under 2% (September 15, 2009)
Fermi massively misses clock targets (November 16, 2009)

Very questionable. Who in their right mind believes that gt300 was only yielding at 2% at one point? That means nvidia has had to improve their yields by 1000% or more in a matter of a few months to get the chip to market.

 

[Soleron]

Those are very, very, sensationalist, misleading, and biased. If you don't like someone/something, you will always find something to slant or bad to say whether it's true, false, fact, opinion, or speculation.




Very questionable. Who in their right mind believes that gt300 was only yielding at 2% at one point? That means nvidia has had to improve their yields by 1000% or more in a matter of a few months to get the chip to market.

Focus on the specific claims, not the wording!

And how do you know he's wrong? We have no chips, no benches, no power figures and no stock figures. Charlie could be right on all of them, or none. It's premature to declare with certainty he's lying.

With the yields thing, he's currently claiming single-digit yields, so let's say 9%. 2% to 9% from A1 to A3 is realistic, that's not 1000% improvement. In his latest article he says Nvidia knows 9% isn't enough to launch, so they won't ship Fermi in quantity and will cut back on production to avoid losses. That's all perfectly consistent with his earlier 2% figure.
And why did you not address the tens of other claims he made and I just listed?

 

[Tempered81]

A look at the Nvidia GT300 architecture (May 14, 2009)

- Whole article focuses on how the chip is focused on GPGPU, many months before Nvidia announce it as such

Here are some pieces before "SemiAccurate" & "Fudzilla" were born (From when Charlie and Fuad both wrote for theINQ):

Nvidia Grabs Stexar Folk (Sept 18, 2006)

-"It looks like Nvidia scored a coup by hiring many if not all of a cohesive design team that hits many of the company's weak spots. The obvious things like STBs and phone chips are what NV would want the Stexar folk for, but a bunch of ex-Intel x86 folk could have some very interesting implications.
No, make that shatteringly huge implications. µ"

Nvidia at Work With Combined CPU and Graphics (Oct 19, 2006)

-"Nvidia's just-announced Portland, Oregon, Design Center, where chip folk are beaving away on 45 nanometre designs.
.....as Nvidia prepares plans to compete with Intel and AMD on the blended graphic and CPU concept."

Nvidia Stexar move turns gun turrets on AMD, Intel (Oct 23, 2006)

-"Where does a team of engineers of this calibre go when they are not really welcome at AMD or Intel? Easy, greener pastures was not a euphemism, they went to Nvidia. What are they doing at Nvidia? Well duh, they are making a CPU."

Nvidia has balls and may circumvent X86 licences (Oct 24, 2006)

-"NVIDIA IS MAKING a CPU, but the only questions are what kind of CPU, and how the heck is it going to do it...
Buy the new Nvidia CPU, it doesn't run Windows and won't run your software or games, but the theoretical graphics power is astounding! ....Nvidia might just end up with an astoundingly good team that will easily beat the best of what is out there, negotiate the maze of patents, and end up winning. The future of high performance computing may be a lighter shade of green"

Nvidia GT300 tapeout soon (April 9, 2009)

-"Nvidia's GT300 is set to tape out in June."

Nvidia GT300 promised in October (April 17, 2009)

-"Nvidia is promising partners will have it in mid-October. If it tapes out like we heard in June, lets just say June 1, that is a mighty tight timeline. Assuming four-week months for brevity, if you hot lot the test batches, you will get first silicon back in eight weeks. That puts testing of A0 silicon starting August 1, and that will take about two weeks. If there are no bugs, or very minor ones, production can start on August 15, and that takes 10-12 weeks. Ten weeks from wafers in brings you to Nov 1, which is slightly after the claimed Oct 15 date. Assuming the AIBs need a few days to build the boards and ship them, you likely won't see them on the shelves until mid-November at the earliest. All of this has the assumption that thing go perfectly, something that we have only heard of once in A0 GPU silicon. If they have to do only one respin, that puts Nvidia out ten more weeks, eight for silicon, and two for testing. Nov 1 now becomes Jan 15. Since Nvidia is 'teh awsum' at making new parts of late, a second respin isn't all that much of a stretch, so it could be out around this time next year."

 

[Soleron]

@Tempered81

Thanks. I did find one Inq article but missed those other two. So he was predicting two respins means April even a year ago. Parts on shelves in volume may yet take to April despite a March launch, so not that far off.

--

As for the CPU claims, he still says they are making a CPU. To make your first x86 CPU has got to be a 5-10 year project. It's no surprise we haven't heard anything yet; they would be very afraid of anyone knowing their intentions on that front and unlike GT300 they don't have to be cooperating with external companies (like TSMC) yet so leaks will be scarcer.

Or they could have canned it. It is an incredibly dangerous and expensive project, even for Nvidia. They could do all that work then have their licensing efforts shot down.