NVIDIA Kepler speculation thread

[Silent_Buddha]

Is it completely out of the question, to have a very well priced GTX 680, if Nvidia is planning to release the GK110 hardware with a GTX 7XX name?

If GK104 can really trade blows with Tahiti, even if it loses, but comes at a more reasonable price (the 1GB GDDR5 edit->less should also help with that), Nvidia could play the good cop for this round, while still maintaining the option to bring 7XX GK110s, with a heftier premium. Everyone in the IT industry is hasty to increase their product's numbers anyway.

They "could" but I find it doubtful considering their past pricing history starting with the ill fated NV1 all the way up to GF1xx.

I'm willing to be pleasantly surprised but I'm not holding my breath. At best I expect it to be priced similarly to 7970. So roughly 550 USD. That's assuming similar performance and no outright lead. If it leads in most programs by a non-insignificant number, then I wouldn't be surprised if it was priced around 600-650 USD.

There's always a chance they might price it similarly to the outgoing GTX 580 at around 500 USD. But I'd still be greatly surprised if that happened.

If they price it lower than 500 USD, then I'm heading out to buy a bunch of lottery tickets.

Although there was the odd blip with the 9800 GTX. But there's quite a few differences there. The card was G92 based like the 8800 GT, and with the 3870x2 out, that put a bit of an upper limit on how they could price it. Plus it wasn't faster than the outgoing 8800 Ultra... So the 9800 GX2 got to be the enthusiast priced card.

Regards,
SB

 

[Rangers]

Not really if the rumors of the die size are true. Nvidia has always had better margins then AMD and that's with bigger chips. Similarly sized chips would exacerbate that further and with AMD's CPU division not exactly lighting the world on fire the last thing they want to do is get in a price war with Nvidia killing their GPU margins.

we dont know AMD's profits nor Nvidias on only (discrete, enthusiast) GPU's. Even if we did we dont know every aspect of them. Countless things go into profit.

I doubt Nvidia magically gets lower prices on every die size, or any, in fact it's illogical to think they do.

I do find it funny how important die size is now again that Nvidia has a small one Predictable though. Before I kept hearing it didn't matter because Nvidia was more profitable with bigger dies...never mind that profit dealt with a million complex factors companywide.

If Nvidia was so profitable with big dies why make any small ones at all? in fact they seem to be doing the opposite, copying AMD's smaller dies strategy with gk104. If you cant beat em join em? Seems to be Nvidias motto vs AMD a lot lately(no hotclock, etc)...

 

[trinibwoy]

Although there was the odd blip with the 9800 GTX. But there's quite a few differences there. The card was G92 based like the 8800 GT, and with the 3870x2 out, that put a bit of an upper limit on how they could price it.

I was going to mention G92 too and it's possible we see something similar this time. Dual GPU cards don't dictate single GPU pricing as they aren't perceived quite the same on the market. The latter have more intrinsic value simply due to less hassle and more reliability.

If Nvidia was so profitable with big dies why make any small ones at all? in fact they seem to be doing the opposite, copying AMD's smaller dies strategy with gk104. If you cant beat em join em? Seems to be Nvidias motto vs AMD a lot lately(no hotclock, etc)...

How much smaller is GK104 than GF104? I didn't realize small dies and no hot-clock were revolutionary features. I guess G80 was the first GPU ever

 

[jimbo75]

If 320mm2 is correct for GK104 it should have ~124 good die per wafer & 70% yield. Tahiti would be ~109 good die per wafer at the same yield. If yields are as low as 50% then there would only be a 10 good-die difference per wafer at those sizes.

Tahiti should be yielding slightly better overall due to having a 3-4 month lead, then you have Nvidia making some noises about yields not being great while AMD are saying that they are meeting customer demand. It's likely that the smaller sized GK104 will not enjoy much of an advantage over Tahiti in respect to total yields over the life of the chips.

If Nvidia has a performance winner here I think they'll do their homework and leave some breathing room for AMD. The last thing they want is AMD dropping Tahiti prices, then being forced to drop Pitcairn prices as well. I think Nvidia will have more to lose out of the midrange/performance price war unless they have a very large performance or die size advantage - neither of which seems likely.

 

[Blazkowicz]

I was going to mention G92 too and it's possible we see something similar this time. Dual GPU cards don't dictate single GPU pricing as they aren't perceived quite the same on the market. The latter have more intrinsic value simply due to less hassle and more reliability.

the dual cards were more practical back then, with high end GPUs eating a lot less power. now a single GF114 eats as much or possibly even more power as the old GX2 and 3870X2 cards.

some normal people did buy the 3870X2 or the 4870X2, they show up for sale sometimes. now nobody buys a 6990 or 590 .

 

[Rangers]

If 320mm2 is correct for GK104 it should have ~124 good die per wafer & 70% yield. Tahiti would be ~109 good die per wafer at the same yield. If yields are as low as 50% then there would only be a 10 good-die difference per wafer at those sizes.

Tahiti should be yielding slightly better overall due to having a 3-4 month lead, then you have Nvidia making some noises about yields not being great while AMD are saying that they are meeting customer demand. It's likely that the smaller sized GK104 will not enjoy much of an advantage over Tahiti in respect to total yields over the life of the chips.

If Nvidia has a performance winner here I think they'll do their homework and leave some breathing room for AMD. The last thing they want is AMD dropping Tahiti prices, then being forced to drop Pitcairn prices as well. I think Nvidia will have more to lose out of the midrange/performance price war unless they have a very large performance or die size advantage - neither of which seems likely.

Cant wait to hear how good it is that Nvidia can rake in all this profit by pricing their small die (important now!), well performing chips, highly If we see GK104 at 499+ or whatever.

After all that kvetching about AMD's terrible , horrible, morally wrong SI pricing, it'll be glorious.

I have to wonder if GK100/110 even exists. Since GK104's called 680. Or if it does exist to be released in a few months, will they call it 780?

 

[xDxD]

http://semiaccurate.com/2012/03/07/tsmc-suddenly-halts-28nm-production/

 

[eastmen]

http://semiaccurate.com/2012/03/07/tsmc-suddenly-halts-28nm-production/

yup 3 week production stopage at TSMC . This isn't good for anyone and I wonder if amd's prices are because of this.

Perhaps they knew that chips would bel imited for quite awhile

 

[AnarchX]

Global Foundries will hopefully be a future alternative to TSMC...
Making AMD and Nvidia dream about 2011 launches and now shutting down fabs in mid-2012.

 

[Topman]

the dual cards were more practical back then, with high end GPUs eating a lot less power. now a single GF114 eats as much or possibly even more power as the old GX2 and 3870X2 cards.

some normal people did buy the 3870X2 or the 4870X2, they show up for sale sometimes. now nobody buys a 6990 or 590 .

Well... Pitcairn is smaller than Barts and a very impressive chip from 'performance x area x power' perspective.

A good candidate for a dual chip card. Unfortunately, i think AMD chooses a dual Tahiti monster.

[]s

 

[Kaotik]

Well... Pitcairn is smaller than Barts and a very impressive chip from 'performance x area x power' perspective.

A good candidate for a dual chip card. Unfortunately, i think AMD chooses a dual Tahiti monster.

[]s

We might still get "7870 X2"'s like we got 6870 X2's from AIBs

 

[silent_guy]

we dont know AMD's profits nor Nvidias on only (discrete, enthusiast) GPU's.

We know a couple of things: Nvidia GPUs typically have higher volumes in the same segment. They've consistently been able to ask higher prices for similar products. It will annoy you terribly, but years of better marketing has resulted in a better brand.

I doubt Nvidia magically gets lower prices on every die size, or any, in fact it's illogical to think they do.

Because there is no such thing as volume based pricing?

Before I kept hearing it didn't matter because Nvidia was more profitable with bigger dies...never mind that profit dealt with a million complex factors companywide.

They were more profitable despite the bigger dies.

If Nvidia was so profitable with big dies why make any small ones at all?

Yes. Why? After all, there is an unlimited, inexhaustible supply of people who are in the market of a $300+ GPU.

in fact they seem to be doing the opposite, copying AMD's smaller dies strategy with gk104. If you cant beat em join em? Seems to be Nvidias motto vs AMD a lot lately(no hotclock, etc)...

Oh, stop that nonsense. It's like my girl complaining that her friend copied her drawing of a cat, except that at least she did draw it. You're just whining on the sidelines due to some misguided sense of fandom. Did AMD copy the scala^H^H^Hnon-VLIW SIMD from Nvidia? The ECC protected caches? And the ability to handle more than 1 triangle per second? Shader cores that don't work with clauses?
And now designing a die that's smaller than your competitor is copying a strategy? Laughable.
This may be a big revelation for you, but in this kind of field, everybody comes up with the same ideas all the time. Sometimes one dismisses an idea while the other doesn't. Sometimes there's no time to implement it. Sometimes ideas are copied. It's one of the accepted ways to improving the state of the art and it's a beautiful thing.

 

[silent_guy]

yup 3 week production stopage at TSMC . This isn't good for anyone and I wonder if amd's prices are because of this.

If true: there goes the theory that only 1 company had issues...

 

[silent_guy]

Tahiti should be yielding slightly better overall due to having a 3-4 month lead, ...

(Head explodes.)

... then you have Nvidia making some noises about yields not being great while AMD are saying that they are meeting customer demand. It's likely that the smaller sized GK104 will not enjoy much of an advantage over Tahiti in respect to total yields over the life of the chips.

I think people are a little to dramatic about what was being said. JHH said during the conf call that gross margins would be a few % lower than usual because yield at TSMC wasn't a high as initially predicted. I know there's a famous journalist and his followers who take this as screaming from the rooftops, but that was not my reading at all. Nvidia has had great gross margins over the last year and it's something that's highly valued by Wall Street. Nvidia has also hinted that they would go even higher. Every time they go down, even a little, it's something that needs an explanation. A lower than expected yield does not have mean disaster, it just means that it's lower. But that's not the kind of material that make people flock to your website...

As for AMD meeting customer demand: that's not only a function of yield but also one of price positioning.

All things equal, a smaller chip will have a better yield than a bigger one. If the difference is 10%, the number of good dies per wafer will be larger than that. But things are not equal: Nvidia has always been more creative in make SKUs to squeeze every last semi-functioning die out their wafers. And we don't know the differences in internal redundancy etc. Nothing that hasn't been said before, though.

 

[boxleitnerb]

^ could be a machine failure though, couldn't it?

 

[DarthShader]

If true: there goes the theory that only 1 company had issues...

yeah, now we know two companies had issues: nVidia and TSMC.

 

[3dilettante]

^ could be a machine failure though, couldn't it?

I thought TSMC had more than one line on that process by now. That would reduce the chances of losing all production.

 

[silent_guy]

I thought TSMC had more than one line on that process by now. That would reduce the chances of losing all production.

That, maybe. And the fact that it's not really a line, but a fab with process stations. With each station having more than 1 machine.

 

[3dilettante]

Having multiple machines would allow for changes to be applied to a subset of machines, which would provide a buffer in case a change turns out to lead to duff chips.

It sounds like TSMC had 28nm in production in one fab.
Perhaps it's a building problem. If important sections in the 28nm process run in a specific section of the building, it could lead to a shut down of the whole process.
Another possibility is that it is a whole fab problem, but the other processes have other fabs to fall back on.
That would show up as a cessation of 28nm, and a drop in throughput for the others.

 

[chavvdarrr]

Global Foundries will hopefully be a future alternative to TSMC...
Making AMD and Nvidia dream about 2011 launches and now shutting down fabs in mid-2012.

Just yesterday, AMD payed hundreds of millions to GF, because they cancelled CPU orders, and wanted to become GF-free :/