Nvidia has serious yield problems with the GT200.

[cegras]

At least they're making educated guesses. You're just applying the 'trust the company! You don't know anything!' principle.

I've worked in a clean room before, with 3 inch silicon wafers making microchannels. Even with only two designs per channel, sometimes things just go wrong. Of course, my developer is much, much thicker and after I do UV exposure the I just wash off the undeveloped stuff, but if I can get errors even at that stage of the process you can imagine how things will go wrong when you're etching many, many more layers at the nanometer scale.

Trust your company all you want; but the thing is if one, just one thing, goes wrong, the whole die is already gone. Especially for a GPU. If nvidia had a couple SP's destroyed, it's no longer a GTX 280. Even worse if the errors are in other, less fault tolerant parts. This ain't like a CPU, where you can disable things with a laser and sell it at a lower price. Nvidia either has a GTX 260 or a 280, and the 280 requires this huge die to be perfect in every way. And it really isn't, considering the low clocks and it's failure to reach a 1 TFLOP output.

[Sr7]

At least they're making educated guesses. You're just applying the 'trust the company! You don't know anything!' principle.

I've worked in a clean room before, with 3 inch silicon wafers making microchannels. Even with only two designs per channel, sometimes things just go wrong. Of course, my developer is much, much thicker and after I do UV exposure the I just wash off the undeveloped stuff, but if I can get errors even at that stage of the process you can imagine how things will go wrong when you're etching many, many more layers at the nanometer scale.

Trust your company all you want; but the thing is if one, just one thing, goes wrong, the whole die is already gone. Especially for a GPU. If nvidia had a couple SP's destroyed, it's no longer a GTX 280. Even worse if the errors are in other, less fault tolerant parts. This ain't like a CPU, where you can disable things with a laser and sell it at a lower price. Nvidia either has a GTX 260 or a 280, and the 280 requires this huge die to be perfect in every way. And it really isn't, considering the low clocks and it's failure to reach a 1 TFLOP output.

WTF are you talking about? I didn't ever say anything about trusting a company. I'm just saying that everyone paints low yields so dramatically when it's really a matter of a few percentage points. It's not a drastic difference between good yields and bad yields. That's *all* I'm saying. Let's not overplay this considering it's all 100% speculation anyway..

[savantu]

No offense, but I find it funny that someone on a forum with an excel spreadsheet for approximating yields thinks they know about yields as though NVIDIA didn't know exactly what they would be dealing with before making their chips. Sorry, but when millions of dollars are at stake, you know these things ahead of time, and you calculate around that. Sure it can fluctuate a bit below what you might expect, but not much.

People making a thread like this seem to think that these guys make wafers and just pray they'll get good yields. When people say good or bad yields ,this isn't a major swing. It's a matter of a percent or two. So it's not the difference between 80 or 20 dies like some seem to believe.

To your surprise , I must inform you that laws of physics tend to apply all across this world and Santa Clara or Taiwain aren't exempted.As 1+1=2 across this planet , so are the formulas for yield calculation based on wafer size , die size and defect density.

We're not stone age and NVIDIA jumped from Star Trek timeframe.

The analysis is rudimentary , but gives a ballpark estimate where their yields are.What we do not know is the defect density in TSMC's 65nm process , but we compared it with the perceived leaders in process technology , Intel and AMD.
We also don't know how redundant GT200 is , but everything suggests most parts can be salvaged and sold as low/mid level.

NVIDIA took a risk with such a large chip ; it could be their architecture at fault , from reviews it looks like it is inefficient compared to R700.

No no no you misunderstood. I'm saying that difference between good and bad yields is a very narrow margin. I'm saying it's not like NVIDIA was expecting 80 and got 20.

I don't think anyone said that NVIDIA expected 80% yields.In fact , nobody mentioned what NVIDIA expected.Even Intel doesn't get 80% with Penryn which is 107mm^2.As for your narrow margin , that's BS.
There's plenty of empirical evidence that suggests otherwise.You can target 40% and get 25%.That's huge

When choices are made for a chip , most of the time the performance of the process it is meant for isn't known.The process invariably turns less than expected performance for a very simple reason : complexity is skyrocketing the smaller you go.Even so , further iterations are expected to improve process performance to planned levels.

AMD expected K10 to achieve 2.2-2.8GHz@95w at 65nm.We all know how it turned out.Are you implying that AMD engineers were idiots and NVIDIA is full of neo-Einsteins that know everything ahead of time ? All the simulations in the world can't replace the cruel reality of tape out.And the real pains starts when you try to mass manufacture the product.

Intel found it the hard way with Prescott ,AMD with K10 , NVIDIA with FX5800 , now again with GT200 , a never ending story.

We simply calculated their yields with the available data.The results are poor , at least compared to ATI , but it was a calculated risk from NVIDIA.Whether their gambit will payoff remains to be seen.Analysts however quickly jumped on this and for good reason.Performance /die size is poor for GTX280 which could make it a flop.

So , what's your point after all ?

[zerazax]

He doesn't have one, he just defends Nvidia whether right or wrong.

Dude, seriously though, you might be right about some things, but you're trying to argue against people who have real world experience in this field and have spent years studying it? Nvidia doesn't always make the right decision! This is why Intel pays top money for fab experts to improve yields even slightly because the difference is HUGE between a few defects per die area and that means huge differences in profits and margins. Laws of physics are laws of physics and when you manufacture off a wafer, often the same procedures are used. I'm sure they did this analysis before hand, and since TSMC doesn't state their defects per area its speculation, but it's probably pretty safe to say they aren't doing better than Intel and even in that best scenario, the yields are still < 40&#37;. A doubling of defects / cm^2 and you can see yields drop into the 10% or so. Things are exponential when you deal with area, so its not just 1-2% differences we're talking about here.

[savantu]

He doesn't have one, he just defends Nvidia whether right or wrong.

...

And you know the funny part ? We did a technical analysis , a statement of findings , not whether Nvidia is right or wrong.

[Xoulz]

WTF are you talking about? I didn't ever say anything about trusting a company. I'm just saying that everyone paints low yields so dramatically when it's really a matter of a few percentage points. It's not a drastic difference between good yields and bad yields. That's *all* I'm saying. Let's not overplay this considering it's all 100% speculation anyway..

It isn't a few % points. We are probably talking in terms of 10% or more. You should read this whole thread.

[motopen1s]

I think Sr7 got rather misunderstood actually guys.


A little offtopic:

As for the laws of physics mentioned twice already - I see the context you guys (savantu & zerazax) are using it in but... To me, the claims sound wrong and in fact, they are very arguable.

Allow me just to quoute one of the authors who would give a good run to a statement that have been used:

"... Indeed, given that the universe and the laws of physics that govern it are also products of the flux, then they, too, must be viewed as habits. Clearly they are habits that are deeply ingrained in the holomovement, but supernormal talents such as immunity to fire indicate that, despite their seeming constancy, at least some of the rules that govern reality can be suspended. This means the laws of physics are not set in stone, but are more like Shainberg's vortices, whirlpools of such vast inertial power that they are as fixed in the holomovement as our own habits and deeply held convictions are fixed in our thoughts..." (c) Holographic Universe by Michael Talbot

Well, one have to read the book in order to understand what is the whole discussion about... In any case - there is planty of literature out there that can provide some insight not only on the laws of physics but much more...

Hard to say it better then Arthur Schopenhauer: "Every man takes the limits of his own vision for the limits of the world."


Its not a 'smart ass' coment of any kind, but just a reminder for everybody including myself that we should rather be a little more careful presenting 'facts' as facts, even when they seem really factual.

As for yelds, in my opinion, everything is even more uncertain then with physics as most of you mentioned here.


peace

[cegras]

Yeah, but to nitpick, we're not really talking about laws of physics. This is a mathematical model calculated based on a data so you can infer or at least predict (not guess) as to what will happen. That's what SR7 doesn't seem to get. He is predicting, not guessing.

Your margin thing is even more pulled out of thin air than his supposed 'guessing'. I have to inform you that 1/93 translates to more than one percent, and the difference between a 'good' and 'bad' yield is not the difference between getting 40 dies out of a wafer versus 41.

[villa1n]

I think Sr7 got rather misunderstood actually guys.


A little offtopic:

As for the laws of physics mentioned twice already - I see the context you guys (savantu & zerazax) are using it in but... To me, the claims sound wrong and in fact, they are very arguable.

Allow me just to quoute one of the authors who would give a good run to a statement that have been used:

"... Indeed, given that the universe and the laws of physics that govern it are also products of the flux, then they, too, must be viewed as habits. Clearly they are habits that are deeply ingrained in the holomovement, but supernormal talents such as immunity to fire indicate that, despite their seeming constancy, at least some of the rules that govern reality can be suspended. This means the laws of physics are not set in stone, but are more like Shainberg's vortices, whirlpools of such vast inertial power that they are as fixed in the holomovement as our own habits and deeply held convictions are fixed in our thoughts..." (c) Holographic Universe by Michael Talbot

Well, one have to read the book in order to understand what is the whole discussion about... In any case - there is planty of literature out there that can provide some insight not only on the laws of physics but much more...

Hard to say it better then Arthur Schopenhauer: "Every man takes the limits of his own vision for the limits of the world."


Its not a 'smart ass' coment of any kind, but just a reminder for everybody including myself that we should rather be a little more careful presenting 'facts' as facts, even when they seem really factual.

As for yelds, in my opinion, everything is even more uncertain then with physics as most of you mentioned here.


peace

LoL. Of course they are not set in stone. There is no such thing as a physical "LAW", as all we are basing those on, is observable events, to test our theory. And what we observe is only a measure of the instrument being used, relative to it.. whether it be the human nervous system, or a volt meter. Granted there are some aspects of physical reality that are pretty damn consistant, lol, but to apply 100% certainty to anything existential is of course absurd.

[Warboy]

This is all you people do isn't it?


You've guys have completely gone off-topic.
This isn't a Confirmed news, I don't care if you have "Sources", No Links means No way to prove it.

And Like My early statement. If News had it's own form of Bloatware, This would be it!

[Papu]

This is all you people do isn't it?


You've guys have completely gone off-topic.
This isn't a Confirmed news, I don't care if you have "Sources", No Links means No way to prove it.

And Like My early statement. If News had it's own form of Bloatware, This would be it!

i guess that makes you Mcaffe?

[K404]

utterly pointless anecdote:

During my time at uni, I was taught semiconductor devices by Prof Asen Aenov. He did a lot of work on Monte-Carlo simulation and as far as im aware, was "The Man" for that and a whole range of other semiconductor phenomenons, including the theory behind reducing tech nodes and all the effects.

His right-hand man in the department was Dr. Jermey Watling- a stastical mathematician who kept Asenovs theory grounded in the real-world probabilities of simulation Vs simulation+ deviation Vs actual Fab.

AFAIK, Asenov never EVER donned a bunnysuit and took part in actual Fab work. (LOL...come to think of it, at the time I was probably better versed in the hands on work of the JWNC/cleanroom protocol than he was, and I was a snotty undergrad)

anyway..... to try and salvage a point- The theory and simulation can be as good as humans can make it, statistics are taken into account for real-world Fab and in a similar way to AMD attempting native quad on 65nm, nV might have pushed their luck too hard this time, maybe partly because the sim didnt transfer to the wafers.

[Nasgul]

I just wanted to have some fun You can take my word for it or not, but I know my source is pretty solid.

No offense but someone starting "some" not credible yet news? Isn't going to cut it, specially when that someone has an ATI avatar.

But I'll just take it as having some fun; seems like there's plenty of GTX2series around here in the U.S., problem and main reason why people are not getting it?
1-Economy not so good
2-Summer time, money towards party and beaches
3-Gas to expensive to trade from a G92 to GTX2series
4-Ultimately, no reason at all to go from G92 to GTX2series for those that have the cash to shell out for.

Just like Ford and GM have stopped the truck division from coming out with new models.......go figure!


__________________________________________

It has been an Nvidia bash-fest since G92 and the huge ripoff it was compared to G80

G92 a huge rip off?

I highly doubt that. I had a G80 and the G80 was the rip off, I paid $575 for the 8800GTX while I just paid $260 for my 9800GTX, yet the 9 series is much better IMO than the 8 series, price and performance. Now? You can get a 9800GTX for $200, certainly not a huge rip off but a great upgrade for someone with a 6 and 7 series card.
Obviously not a good upgrade with those with 8 series cards but.....

[xVeinx]

I'm sensing a clash of realist and anti-realist philosophies of science . Even when making more broad, sweeping generalizations, whether or not we believe the physical laws to be entirely constant, I think it safe to say that yield problems can and do exist. I also think it safe to say that large, complex cores will little cache involved are more susceptible to defects at a larger process size (even if it's merely "perception" in observation). There is only so much binning you can do with these cores (until the 280GSO or whatever); its either 280GTX or 260 right now, nothing more. One has to assume that the stream processors are fully functional to begin with, and even then, able to clock to at least the stated base speeds (which are usually set based on the best scenario with turnout and competition). It's not rocket science-the cores are pricey even with great yields. There is a reason they are charging as much as they are for these cards and not able to reduce prices like they were with the 9800GTX....

[Nasgul]

4870 randomly bit the dust in less than 24 hours - nice QA.

[zerazax]

What the hell does that have to do with this thread? Besides, those guys modded their BIOS so thats not something QA has to do with it.

[Warboy]

i guess that makes you Mcaffe?

Aww. I wanted to be Norton.

[K404]

resource hog and over-rated?

JK

[Papu]

4870 randomly bit the dust in less than 24 hours - nice QA.

you missed the part where one person flashed their bios's , can i complain to intel if their chip wont boot at 1.7V?
things randomly dying happens to every piece of hardware , nvidia are not immune :P

It looks like Nvidia is shaving it pretty close, as far as GTX280 GPU is binning is concerned. I say this because my GTX280 just died! - after less than an hour of 3DMark06 looped testing. And this was on my open test jig, with max air cooling (fan duty cycle was set to 70%).

source

at least the ati cards last longer

[villa1n]

you missed the part where they flashed their bios's , can i complain to intel if their chip wont boot at 1.7V?

LOL. no doubt. Sounds like they raised their clocks higher than the current voltage they were running would support.

[xVeinx]

4870 randomly bit the dust in less than 24 hours - nice QA.

They screwed up when they flashed them, but somehow that's ATI's fault. Brilliant analysis...

[cegras]

No offense but someone starting "some" not credible yet news? Isn't going to cut it, specially when that someone has an ATI avatar.

But I'll just take it as having some fun; seems like there's plenty of GTX2series around here in the U.S., problem and main reason why people are not getting it?
1-Economy not so good
2-Summer time, money towards party and beaches
3-Gas to expensive to trade from a G92 to GTX2series
4-Ultimately, no reason at all to go from G92 to GTX2series for those that have the cash to shell out for.

Just like Ford and GM have stopped the truck division from coming out with new models.......go figure!


__________________________________________

G92 a huge rip off?

I highly doubt that. I had a G80 and the G80 was the rip off, I paid $575 for the 8800GTX while I just paid $260 for my 9800GTX, yet the 9 series is much better IMO than the 8 series, price and performance. Now? You can get a 9800GTX for $200, certainly not a huge rip off but a great upgrade for someone with a 6 and 7 series card.
Obviously not a good upgrade with those with 8 series cards but.....

I see you bought the 9800 GTX after seeing it perform neck in neck with your 8800 GTX. Congrats, you have been milked by nvidia .. it's okay to feel a bit butthurt.

Again, while it's okay to express skepticism, outright outrage against Alexio is a bit over the top. Especially since if we can do some basic calculations which pretty much back it up.

Everyone should have predicted that a 1.4 billion transistor die would have yield problems. Everyone.

[K404]

Everyone should have predicted that a 1.4 billion transistor die would have yield problems. Everyone.

QFT!

[Nasgul]

I see you bought the 9800 GTX after seeing it perform neck in neck with your 8800 GTX. Congrats, you have been milked by nvidia .. it's okay to feel a bit butthurt.

Had the 8800 for only 6 month, sold it for $525 and put the 6800 GS I had laying around. It's not like I sold my 8800 for $100 just to trade up. I'm glad I waited this long and not rush into getting some HD3 series crappola.

Again, while it's okay to express skepticism, outright outrage against Alexio is a bit over the top. Especially since if we can do some basic calculations which pretty much back it up.

Calculate with something to back it up? Calculate what and what's there to back it up with? Other than someone "having fun".

Everyone should have predicted that a 1.4 billion transistor die would have yield problems. Everyone.

Everyone? Is that why so many don't think this is "credible" news?

calculation and predictions IS NOT exactly knowing. Again, where is such thing as nVidia is having issues "actually"? Care to post a link or "everyone" should know because "someone" is saying so OR........

We should know because we know how to "calculate" and "Predit"?

Which one will it be?

[K404]

Statistically the probability of device failure increases with transistor count. A system is only as strong as its weakest component, so its no big jump to say that a 1.4B transistor chip is very ambitious and more likely to fail than a chip with a smaller transistor count, even taking into account different parameters between large and small chips.