Nvidia GT300 yields are under 2%

[zanzabar]

while the yields seam low, this is the should be the 1st batch, and it dosnt say what they wanted. while it sounds like and i would guess is BS, they could have wanted a fully working chip in 40nm and that clocks vary high for a top tier, then that would make sense that they have a reported 2% yeild. NV basses the gpu tiers or getting fully working, and partially working chips, so lets say that they got 2% of the top chip the 1st try that seams good to me.

charlie spins things so if u look at it how i suggest as the logical way for it to be true then it could make sense, there could be the reported 30% or whatever it was then the top chip gets 2%.

[JohnJohn]

If you google "Charlie Demerjian RV870 (R800, RV8xx, etc) shaders (stream processors, SP's, etc)" you'll find that he has claimed 1200, 960, 2000, 1400, and 1600 over the past year.

Sorry man, i googled and the only one i found was the one about the 1600 shaders, Can you please back Up what you said?

I also have been reading charlie for a while, and one thing i can tell is that he isn't so stupid of going through life throwing specifications numbers, because numbers are the easiest ones to debunk

[eric66]

well as long as nvidia is silent this rumour nuber will grow so nvidia better show something

[Humminn55]

From a member at Anandtech:

Can it be that bad? Sure, it can always be zero.

Let's just assume ALL of Charlie's numbers and sources are 100% correct...its four wafers.

Getting low yields on four wafers is not exactly uncommon. And it especially comes as no surprise for a hot lot as typically hot lots have nearly all the inline inspection metrology steps skipped in order to reduce the cycle-time all the more.

Those inline inspections are present in the flow for standard priority wip for a reason, related to both yield (reworks and cleanups) as well as cost reduction (eliminate known dead wip earlier in the flow).

I really pity anyone who is wasting their time attempting to extrapolate the future health of an entire product lineup based on tentative results from four hot-lotted wafers. That's not a put down to anyone who is actually doing just that, including Charlie, its an honest empathetic response I have for them because they really are wasting their time chasing after something with error bars so wide they can't see the ends of the whiskers from where they stand at the moment.

Now if we were talking about results averaged from say 6-8 lots and a minimum of 100-200 wafers ran thru the fab at standard priority (i.e. with all the standard yield enhancement options at play) then I'd be more inclined to start divining something from the remnants of the tea leaves here.

But just four wafers? Much ado about nothing at the moment, even IF all of the claimed details themselves are true.

This would have been far more interesting had the yields on those four wafers came back as 60% or 80%, again not that such yield numbers could be used to say anything about the average or the stdev of the yield distribution but it would speak to process capability and where there is proven capability there is an established pathway to moving the mean of the distribution to that yield territory.

But getting zero, or near-zero, yield is the so-called trivial result, it says almost nothing about process yield to get four wafers at zero yield. All it takes is one poorly performing machine during one process step and you get four wafers with yield killing particles spewed on them.

[xlink]

Ouch! Only 7 out of 416 chips were useable? I hope things turn out better soon...Maybe Nvidia went too big this time around?

they went too big last time around, do you honestly think this chip is any smaller than their previous chip?

that said, I expect yields to be way higher by launch in 6 months and expect the current number of "mostly usable" chips to be much higher than 2%

[Solus Corvus]

I'm willing to bet it's not true. And it's bad news if it is.

[Beefy22]

Getting quite tired of that as well. As far as I can tell BenchZowner is just an nvidia fanboy and a troll. No disrespect to you BenchZowner but trying to prove to others that you have more information than everyone else without producing anything doesn't help your credibility.

...

Out of curiosity, what do you consider disrespectful?

Anyway, perhaps BenchZowner is under NDA? 2% yields seems almost unfathomable.

[Mechromancer]

...

Out of curiosity, what do you consider disrespectful?

Anyway, perhaps BenchZowner is under NDA? 2% yields seems almost unfathomable.

LOL I agree. That is like saying, "I mean no disrespect, but your mother is a fat, ugly, whore."

We should go back to how it used to be at a time when you knew the line between respect and disrespect because when you crossed it, you quickly lost your life at the tip of a sword. The good ole days....

[zerazax]

Anyway, perhaps BenchZowner is under NDA? 2% yields seems almost unfathomable.

If you're really under NDA, you wouldn't (or shouldn't) make baseless claims like that...

And it's not like this is anything new he's been doing anyways. He said the same thing about the GT200s, the G92s, etc. vague comments that seemed to hype it up but it's not hard to make vague comments while pretending to be under an NDA and pretending you're in the know

[MirageSys]

Demerjian is the Uwe Boll of tech reporters.

More like the Ebola. I think he must have a phd in proctology because I don't know how he pulls so much stuff out his ass.

[situman]

Well they can always transform those into mobile chips or low to mid level cards if these rumours are true. I mean can a chip can be completely useless? Maybe not up to spec, but all are completely useless? So its can't be all bad. There just wont be that many $700 gpus that's all.

[zanzabar]

Well they can always transform those into mobile chips or low to mid level cards if these rumours are true. I mean can a chip can be completely useless? Maybe not up to spec, but all are completely useless? So its can't be all bad. There just wont be that many $700 gpus that's all.

they just have more truck sized key chains

[Utnorris]

Let's pray it's not remotely true. We the consumer need the competition otherwise prices will remain high. I hope Nvidia delivers in October, it would mean the HD5870 will be in the $300 price range quick assuming the GT300 beats it. Remember how expensive Nvidia's top cards were a few years back when ATI couldn't compete, we do not want the opposite to happen.

[labs23]

Let's pray it's not remotely true. We the consumer need the competition otherwise prices will remain high. I hope Nvidia delivers in October, it would mean the HD5870 will be in the $300 price range quick assuming the GT300 beats it. Remember how expensive Nvidia's top cards were a few years back when ATI couldn't compete, we do not want the opposite to happen.

Most of us want it that way. But I guess we'll hope for the best to happen for nVidia at the same time expect the worse. I mean October for them is even a big question.

[003]

they went too big last time around, do you honestly think this chip is any smaller than their previous chip?

No they didn't. Plenty of silicon chips are as big or bigger than GT200. People just made a big deal because RV770 was comparatively small. G300 is smaller than GT200 but larger than GT200b.

Larrabee is supposed to be bigger than GT200.

[gumballguy]

#1- Charlie is NOT the source. A different source leaked some info and a very good informant even posted up the numbers Charlie was going to quote yesterday. This informant quoted similar troubles but slightly higher yields, still under 10%.

Why would anyone tell charlie anything? Or perhaps more to the point, Charlie exclusively?

they went too big last time around, do you honestly think this chip is any smaller than their previous chip?

Nope. Maybe they'll think different by GT500. I think the successsors are too far into design now to change so soon. However, they might well scrap some of those and fastrack designs based on smaller chips.

From a member at Anandtech:

Interesting post. Why would anyone ham it up to Charlie if that was the case?

More like the Ebola.

Uwe Boll and the Ebola virus are the same thing, disregarding that one pretends to be sentient.

---

At the end of the day, I think it's Charlie being Charlie. I put absolutely no stock in this rubbish at all. You can argue whether the info was inadequate (based on the anandtech post) or it's just Charlie's NV hate, but either way I'm calling it rubbish.

And I will probably buy ATI this time so no, not a fanboy :p

[bigKr33]

I really really hope this isn't true. I'm not biased towards amd or nvidia, but I was really looking forward to the gt300.

[OBR]

thats totally bull*, we all will see next week .... check news from San Jose

[LordEC911]

thats totally bull*, we all will see next week .... check news from San Jose

Ahhh... You mean the Tegra, PhysX and CUDA power point slide show?

[tajoh111]

thats totally bull*, we all will see next week .... check news from San Jose

Ahhh... You mean the Tegra, PhysX and CUDA power point slide show?

Why don't you make a signature bet with each other. That way OBR can be called out. I don't know what to expect anymore, I just don't trust the rumors.

However I can imagine NV releasing, even if it a single card benchmarks of their card when the 5870 releases as it does one or both of two things. It stops people from buying a 5870 or it makes AMD lower the price of their card. The only reason I think NV is going to release something fast is because the chip is so big and NV I can imagine doesn't want to screw around with larrabee and 5870.

[bigKr33]

thats totally bull*, we all will see next week .... check news from San Jose

San Jose? Thats the gpu technology conference isn't it? What sort of things take place there anyway.

[saaya]

even if its true, it doesnt really mean they will only have 2% yields in mass production... these were only samples
and we dont know what is defined as fully working and whats defined as broken... broken might as well mean that it cant reach the clockspeeds they aimed for, or runs hotter than they aimed for, or that some part of the gpu needs to be disabled, but it still works, ie 260s...

7 working parts is pretty bad for the first sample batch, but it still beats 0 which does happen as well...
so they at least have working parts they can now work with and test, figure out why yields were so bad and adjust the masks or design...
def not good news, if true, but not a major catastrophe id say... the launch is still targeted for q1 which means its around half a year away...

[wiak]

what are you guys whining about?
2003 NVIDIA Geforce FX was a train wreck
2007 ATi Radeon HD 2900 XT was a train wreck
2009 NVIDIA Geforce GTX 300 might be a train wreck

makes sense

[bigKr33]

what are you guys whining about?
2003 NVIDIA Geforce FX was a train wreck
2007 ATi Radeon HD 2900 XT was a train wreck
2009 NVIDIA Geforce GTX 300 might be a train wreck



makes sense

Interms of sales meaning numbers sold, or yields?

[Smartidiot89]

Ok lets imagine Charlie is true about this article.

Does anyone remember the whole "Risk Wafer" deal so that Nvidia could get some GT300 out before 2010? Couldn't these be the product of those Risk Wafers?