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

[GodisanAtheist]

http://www.semiaccurate.com/2010/02/17/nvidias-fermigtx480-broken-and-unfixable/

-Yeah, its Charlie. These are always good for a debate, but keep the gloves above the belt. ATI users who feel that electric thrill of bad things happening to Nvidia, remember that your 5xxx series is going to remain a higher price without competition. Nvidia users, just cause its Charlie and he's talking shit again doesn't mean its certainly not true.

Given the recent Rv870 article from Anand and the nuggets of info regarding TSMC's 40nm process along with ATI & Nvidia's respective design philosophies, this is one of Charlie's *ahem* less outlandish articles. It is, however, very possible that he simply read anand's article and extrapolated this whole thing out of his ass.

Bon apetite!

 

[ScorcherDarkly]

I really hope that article isn't accurate.

 

[tviceman]

Every time I hear about a new Charlie article bashing nvidia, I roll my eyes.... I dislike the guy and his very biased opinions.


That said, to be this close to launching Fermi and having that much information (whether true or false) does not bode well. Either nvidia is having someone slip him old/bad news or Fermi is going to be a very, very limited run card.

 

[WelshBloke]

So we get to find out if he's semiaccurate, slightly accurate or completely full of shit. D:

 

[cbn]

One interesting thing I learned from the RV870 article was that ATI is better at process technology


With ATI claiming to launch a new set of chips by the end of the year could we see Intel try to buy out Nvidia. Even though Nvidia says it makes much more money on workstation boards could Intel's superior process technology make this more attractive?

P.S. I read in Wikipedia that the Intel owned Havok decided not to develop a Physics based graphics engine. Could a purchase by Intel therefore pave the way for making CUDA a graphics standard?

 

[Tempered81]

Whoa, crazy article by SA. NV is still going to launch the 470 & 480 in the next 4-6 weeks. I sure hope the 480 model is a ~750mhz 512 shader chip. If it's ~600mhz + 448 cores, it'll be weak. So the shipping product is GF100-A3?

He does make things seem really grim for NV, however, there is no doubt in my mind that GF100 cards will be totally kick-ass. Sure, they may be hot, power-hungry, defected, money-losing, unmanufacturable, late, shitty and whatever else he wants to call them. But they will be fast as shit in games. And if those channel pricing rumours from Tweakers are true, then they might possibly be cheaper as well. Well, cheaper than with Nvidia's historic top end pricing model.

 

[ScorcherDarkly]

Sure, they may be hot, power-hungry, defected, money-losing, unmanufacturable, late, shitty and whatever else he wants to call them. But they will be fast as shit in games.

Who cares how fast they are if they only make 10,000 of them?

 

[MagickMan]

The man is a f*cking idiot. Why does anyone even bother reading that stupid shit?

I own both NV and ATI cards, and I don't want either to go under.

 

[cbn]

Who cares how fast they are if they only make 10,000 of them?

If that is true, those 10,000 would probably only make it to workstations?

Jen-Hsun Huang has claimed in the past 2/3 of his profits (but only 1/3 of his gross revenue) comes from the HPC market.

 

[Shilohen]

So we get to find out if he's semiaccurate, slightly accurate or completely full of shit. D:

This. We're only 6 week away to know if he is correct from the performance point of view. It's going to take a bit longer to know if the 10k chips figure it true.

Nevertheless, it'll be interesting to come back to this thread at the end of March. So far, Charlie as been pretty accurate with his Fermi information, let sure hope he's not this time however, or we could see at worst another shameless MSRP increase from ATI or at best stagnating prices.

 

[Rebel44]

Why does anyone even bother reading that stupid shit?

Because so far he was mostly correct about Fermi so far.

My opinion: If everything was OK for Fermi and its performance was 50%+ over 5870 NV would leaked performance info looooong time ago. Their silence just fuel rumors that its going to be epic fail just like NV30 and R600.

 

[Lonyo]

Well all of what he says makes sense, to a point.
The true extent of the problems may be questionable, but that the problems exist is reasonably certain, and the causes of them are the same as those Anand has recounted.

It wouldn't be unreasonable for this article to be at least semi-accurate, since it's nothing entirely new (other than the catastrophic performance/yield claims which are not something we are likely to ever know the truth of).

Facts:
Fermi die is larger than Cypress.
AMD gained experience with the whole HD4770 thing
There is no 40nm G200 based card
TSMC 40nm process has been problematic
Fermi has been delayed

Add it all up and it doesn't look, from the outside, particularly rosy for NV. When AMD who have successfully launched a product talk about the fact that they had to work around problems using their experience, experience NV doesn't have, you can argue that it seems reasonable to think either NV/TSMC have magic powers, or they are going to have problems, or at least a product which will have worse yield than AMD products.

 

[JAG87]

I read "sources told semiaccurate" about 4 times.

I'd really like to know who the fuck are these sources that spread information like this apart from nvidia themselves.

This man just dreams something, wakes up, and puts all his deranged thoughts in writing on a shitty website that has no reputation at all.

 

[cbn]

Well all of what he says makes sense, to a point.
The true extent of the problems may be questionable, but that the problems exist is reasonably certain, and the causes of them are the same as those Anand has recounted.

It wouldn't be unreasonable for this article to be at least semi-accurate, since it's nothing entirely new (other than the catastrophic performance/yield claims which are not something we are likely to ever know the truth of).

Facts:
Fermi die is larger than Cypress.
AMD gained experience with the whole HD4770 thing
There is no 40nm G200 based card
TSMC 40nm process has been problematic
Fermi has been delayed

Add it all up and it doesn't look, from the outside, particularly rosy for NV. When AMD who have successfully launched a product talk about the fact that they had to work around problems using their experience, experience NV doesn't have, you can argue that it seems reasonable to think either NV/TSMC have magic powers, or they are going to have problems, or at least a product which will have worse yield than AMD products.

If any of that semi-accurate article is true, rest assured Nvidia is working on some new patents to get around this process technology problem they have.

Linked and Synchronized small GPUs? Pooled rather than "mirrored" VRAM?

 

[WelshBloke]

I read "sources told semiaccurate" about 4 times.

I'd really like to know who the fuck are these sources that spread information like this apart from nvidia themselves.

This man just dreams something, wakes up, and puts all his deranged thoughts in writing on a shitty website that has no reputation at all.

Its either people at TSMC or the card manufacturers (EVGA, XFX etc), my bet would be TSMC as apparently NV has been giving them a bit of a hard time.

 

[zerocool84]

We all know that if Nvidia really was ready to release so soon, they would have released more info to keep people like me from jumping ship.

 

[Lonyo]

I read "sources told semiaccurate" about 4 times.

I'd really like to know who the fuck are these sources that spread information like this apart from nvidia themselves.

This man just dreams something, wakes up, and puts all his deranged thoughts in writing on a shitty website that has no reputation at all.

Guess who else wants to know who these sources are.
That's right, NV. Hence they remain as unnamed sources, because Charlie doesn't want to lose them, and they don't want to get in trouble.

 

[Tempered81]

Who cares how fast they are if they only make 10,000 of them?

Well if they make 10000 and you get one of them, and it's really fast, you should be happy. No?

 

[SolMiester]

LOL, what company would press ahead with a single figure % yield on the 3rd re-spin?.....none.
Thats why the article is crap, even NV wouldnt waste the money!

 

[Rebel44]

LOL, what company would press ahead with a single figure % yield on the 3rd re-spin?.....none.
Thats why the article is crap, even NV wouldnt waste the money!

I agree that Fermi yield is likely higher but its IMO very likely its still under 20% and that with such delay it will be seen as epic fail unless its significantly faster than 5870 and with reasonable price and availability.

 

[ScorcherDarkly]

Well if they make 10000 and you get one of them, and it's really fast, you should be happy. No?

Sure, I'd be happy. So would ATi. The rest of the GPU-purchasing populace, not so much.

 

[Keysplayr]

I read "sources told semiaccurate" about 4 times.

I'd really like to know who the fuck are these sources that spread information like this apart from nvidia themselves.

This man just dreams something, wakes up, and puts all his deranged thoughts in writing on a shitty website that has no reputation at all.

Then why does everyone sit around talking about him? He knows how to get attention. Just like the ancient Romans in the coliseum. Everybody loves a good bloodbath. Seems to hold true even today under the guise of civilization.

 

[Tab]

It's a well-written article, extremely well done. If what he says is true I'd bet we won't see any GPUs from nvidia going mainstream this year. Until then, we can only wait.

EDIT - This actually wouldn't be surprising, given the fact nVidia has cut down on R&D and other investments. Read SAs article about cutting costs and investments.

 

[AyashiKaibutsu]

http://www.semiaccurate.com/2010/02/17/nvidias-fermigtx480-broken-and-unfixable/

Nvidia users, just cause its Charlie and he's talking shit again doesn't mean its certainly not true.

It does mean that it's completely made up fantasy. Whether what he says and what actually happens are the same is completely by accident when they coincide; any person here or on any subforum on anand could make just as accurate predictions as charlie has.

 

[SHAQ]

We will know in a month. Don't wait to buy one if you want one.