What comes after Fermi?

W

WabeWalker

Supreme [H]ardness
Joined
Sep 3, 2007
Messages
4,508
And when?

Any chance that we'll see something this year? My GTX 480 is flying, still, at 2560x1440, but extra frames are always nice.

Anybody here got a 480 sli setup - any troubles with that?
 
all speculation. May be a dual chip card in a few months. No new architecture until 2012.
 
WabeWalker said:
And when?

Any chance that we'll see something this year? My GTX 480 is flying, still, at 2560x1440, but extra frames are always nice.

Anybody here got a 480 sli setup - any troubles with that?
Click to expand...
Probably nothing until late this year. The next architecture is called Kepler and will be 28nm. That's all that is really known as of right now AFAIK.
 
NKD said:
all speculation. May be a dual chip card in a few months. No new architecture until 2012.
Click to expand...
I believe NVIDIA has said Kepler will be available in 2011 but I won't be surprised if nothing ships until 2012.
 
28nm is what comes next, and that's a long way off. I'd hazard a guess at the first 28nm cards arriving between November 2011 and March 2012, at the earliest, probably later.
 
Rebrand/die shrink, then Kepler, IMO. nVidia has already stepped the in crap pile twice, for trying new arch + new process at the same time.
 
yes stuff will be ready at the end of this year, but they have to be able to manufacture it. From what everyone has said 28nm wont really take off for GPU until Q1 2012, and in best case scenario Jan or Feb. Yea may be at the end of this year if the reports are wrong. Ati was ready for 28nm with norhtern islands but didn't happen, thanks to TSMC. I wont be surprised since they always seem to mess up on every new process, they are not intel.
 
jeremyshaw said:
Rebrand/die shrink, then Kepler, IMO. nVidia has already stepped the in crap pile twice, for trying new arch + new process at the same time.
Click to expand...

So has ATI so what's your point ?

Anyway, there's no die shrink before Kepler. Kepler is the next thing coming.
 
Silus said:
So has ATI so what's your point ?

Anyway, there's no die shrink before Kepler. Kepler is the next thing coming.
Click to expand...


theres no die shrink because TSMC isn't doing 32nm they skipped and went to 28nm. seeing as how long it took them just to get 40nm correct i doubt a year and half is enough time for them to perfect 28nm. so early keplar will probably be crap. on the other hand GF's been testing 28nm fab for over a year now. so it's probably going to come down to another race between who gets 28nm out first.
 
sirmonkey1985 said:
theres no die shrink because TSMC isn't doing 32nm they skipped and went to 28nm. seeing as how long it took them just to get 40nm correct i doubt a year and half is enough time for them to perfect 28nm. so early keplar will probably be crap. on the other hand GF's been testing 28nm fab for over a year now. so it's probably going to come down to another race between who gets 28nm out first.
Click to expand...

Different processes, different challenges. You can't compare one with the other...
 
sirmonkey1985 said:
theres no die shrink because TSMC isn't doing 32nm they skipped and went to 28nm. seeing as how long it took them just to get 40nm correct i doubt a year and half is enough time for them to perfect 28nm. so early keplar will probably be crap. on the other hand GF's been testing 28nm fab for over a year now. so it's probably going to come down to another race between who gets 28nm out first.
Click to expand...
I think a lot of the problems were with architectural design challenges too, so I'd assume that what ATI and NVIDIA learned on the last die shrinks should help with moving to 28nm.
 
Rizen said:
I think a lot of the problems were with architectural design challenges too, so I'd assume that what ATI and NVIDIA learned on the last die shrinks should help with moving to 28nm.
Click to expand...

well its a lot of things, nvidia did 0 testing on the 40nm process before hand and what testing they did had nothing to do with fermi except the fact that TSMC could do a 40nm process even though it ran hotter and was more power hungry then the 55nm and 65 processes (GT 2xx line). but the other big thing was the yield problems on TSMC's side which effected both sides. now that AMD wont be using TSMC for their 28nm GPU's maybe the problems will go away since they are no longer having to produce 2 totally different architectures. but i still have my doubts due to nvidia continuing to use these over complicated architectures and over sized cores and i don't see that trend changing any time soon. g92 was our glimmer of hope that maybe they changed their ways but that died a swift death with the G200.
 
sirmonkey1985 said:
well its a lot of things, nvidia did 0 testing on the 40nm process before hand and what testing they did had nothing to do with fermi except the fact that TSMC could do a 40nm process even though it ran hotter and was more power hungry then the 55nm and 65 processes (GT 2xx line). but the other big thing was the yield problems on TSMC's side which effected both sides. now that AMD wont be using TSMC for their 28nm GPU's maybe the problems will go away since they are no longer having to produce 2 totally different architectures. but i still have my doubts due to nvidia continuing to use these over complicated architectures and over sized cores and i don't see that trend changing any time soon. g92 was our glimmer of hope that maybe they changed their ways but that died a swift death with the G200.
Click to expand...
The GF110 is a pretty damn good design given the overall size of the GPU. I don't think there is anything wrong with producing a large die and a complicated architecture, as long as they are able to produce it reliably and with a profit. We'll have to see. I would guess NVIDIA is loathe to repeat the GF100 disaster.
 
jeremyshaw said:
when? so what it your point, really?
Click to expand...

ATI screwed up R600 and R520, R600 being the biggest failure. R520 was very late and short lived (3 months), but was at least on par or a bit faster in some games, when compared with the competition.

The point was quite clear...Both companies had problems when going to a new architecture, on a few occasions, but that won't stop them from going forward with a new process for it.
 
Brashen said:
So now with the die shrink will they introduce dual chip on that?
Click to expand...

Dual chip is rumored to be ready and waiting for AMD's Antilles.

And again, there is no die shrink. Kepler is the next GPU from NVIDIA and it will be on 28nm. It won't be a new architecture, but rather an evolution of Fermi.
 
sirmonkey1985 said:
well its a lot of things, nvidia did 0 testing on the 40nm process before hand and what testing they did had nothing to do with fermi
Click to expand...

0 testing...Really ? Proof ?
I just hope that it's not the type of proof you provided when you claimed the 4-wide SIMDs in Cayman would be super awesome...

sirmonkey1985 said:
but i still have my doubts due to nvidia continuing to use these over complicated architectures and over sized cores and i don't see that trend changing any time soon. g92 was our glimmer of hope that maybe they changed their ways but that died a swift death with the G200.
Click to expand...

Why would they change a strategy that's working out for them ?
A GPU that is great for graphics and GPGPU, which targets three markets - consumer, professional and HPC - two of them with extremely high profit margins, can't possible be better and their finances and latest 10K show that quite clearly.
 
Silus said:
Kepler is the next GPU from NVIDIA and it will be on 28nm. It won't be a new architecture, but rather an evolution of Fermi.
Click to expand...

All I care about is the performance. It better be at least 50% faster than a GTX 580. None of this barely 20% better BS.
 
Die-shrinks always bring about huge increases in performance, assuming they come with a new product, otherwise they just bring about big efficiency improvements. The GTX580 was only a mere 20% better than the GTX480 because it was a 40nm card just like the 480 was, and since the 480 already used 300W, they couldn't gain anything by making it bigger.
AMD have made the GPUs bigger but still only gained 20% because they've made the GPUs far more suited to heavier workloads instead.

HD5870 -> HD6970 (40nm->40nm) 17-20% improvement, 210W vs 190W.
HD4890 -> HD5870 (55nm->40nm) 60% improvement, 190W vs 190W.
HD3870 -> HD4850 (55nm->55nm) 80% improvement, 110W vs 106W (A rare case, simply due to the badness of the HD2/3 series)
HD2900 Pro->HD3870 (80nm->55nm) 30% improvement, 106W vs 180W (As above)

GTX480 -> GTX580 (40nm->40nm) 20% improvement, 295W vs 295W (TDPs aren't accurate for these cards)
GTX285 -> GTX480 (55nm->40nm) 55% improvement, 295W vs 204W (A bit of Fermi-fail here, but nonetheless performance gains)
GTX280 -> GTX285 (65nm->55nm) 10% improvement, 204W vs 236W
 
You must log in or register to reply here.
Back
Top