Who will deliver the first 1TFLOPs retail GPU?

[Acert93]

The subject line says it all: Who will be the first GPU maker to ship a single chip GPU solution with over 1TFLOPs of programmable performance? NV, ATI/AMD, Intel, or other?

And a bonus speculation question: WHEN?

 

[Ailuros]

On paper as in a theoretical maximum or a real time sustained value?

 

[icecold1983]

ati imo

 

[Acert93]

On paper as in a theoretical maximum or a real time sustained value?

Paper 32bit programmable, as that seems to be the number ATI & NV release (as well as STI, Intel, etc). Obviously realworld utilization in non-trivial situations is a much more difficult measure. We could argue all day about what is a non-trivial app and such and it could be months and months before we would ever get confirmed sustained performance. e.g. What is G80's sustained performance in a realworld scenario? Those sort of numbers for any GPU have been hard to come by.

I thought up the question mainly because of the press Intel has received over Terrascale (and in the past the BE) and much like the "first to 1GHz" was good press, I wouldn't doubt the first GPU to reach 1TFLOPs would also be press worthy (read: hyped). And the reality that GPUs are closing in on paper-1TFLOPs in the next 12-24 months so now is the time to speculate As they begin to be used for some general purpose roles and begin competing in some areas with other chips attaining this level of performance (even if in limited scenarios) is interesting.

Not a perfect question and criteria, but the easiest to discuss as long as everyone realizes that 1TFLOPs of paper performance doesn't directly equate to 1TFLOPs of sustained realworld performance (and that flops themselves don't say much for flexibility or utilization; a GPU with lower flops could have higher thoroughput due to a better design).

 

[Ailuros]

Albeit I figure you're asking about single GPUs, you get that theoretical value already with a 8800GTX@SLi system (1036 GFLOPs/s) (ok ok don't throw stuff at me LOL).

 

[Acert93]

Albeit I figure you're asking about single GPUs, you get that theoretical value already with a 8800GTX@SLi system (1036 GFLOPs/s) (ok ok don't throw stuff at me LOL).

The subject line says it all: Who will be the first GPU maker to ship a single chip GPU solution with over 1TFLOPs of programmable performance? NV, ATI/AMD, Intel, or other?

The reason it isn't in the poll is because there is a limit on the number of letters in a poll question

Maybe I should go back and emphasize that in the text in the OP seeing as people probably are only going to read the question and knowing the rowdy tongue in cheek B3D sorts you can be certain that where there is one there are others!

 

[Inane_Dork]

Ati/Amd because their chips have, since the 8500, pushed shader processors ahead of the curve.

Intel won't have a high end part in time to beat either Nvidia or Ati.

Nvidia's part will be a generally better-performing part than the 1 TFLOP Ati part mostly because "performance" will be measured in existing games.

When? No more than 2 years from now.

 

[Ailuros]

Seriously now I voted for NVIDIA because so far they've been consistantly ahead in their release cycles. A hypothetical refresh with 10 clusters, a healthy clock domain boost and dual issue MADD should do the trick. I have a hard time seeing from Intel anything relevant that early.

 

[pjbliverpool]

My moneys on NV with G90 in Q1 next year.

 

[Corwin_B]

Rampage would have done that.

Seriously, though, my money is on AMD.

 

[Davros]

on paper (literaly) bitBoys - i win

 

[epicstruggle]

Amd, before the end of the year.

 

[_xxx_]

nV, next year or end of this year.

 

[Power_man]

ATi of course .The GeForce 7800 GTX have ~198 Gflops (July 2005);The Radeon X1900 XTX have ~375 Gflops (January 2006);The GeForce 8800GTX have ~520 Gflops (November 2006).

I think the future will be:The Radeon X2800 series may be will have ~550/~600 Gflops (March 2007) ; The GeForce 8900 series may be will have ~620/650 Gflops (Q2 2007); The Radeon X2900 series may be will have ~670/680 or ~700 Gflops (Q3 2007) ; The G90 may be will have ~~800/~~900 Gflops (No Date) ; The R700 may be will have ~1Tflops (or R780).
For me,the R700 will be first in world GPU with 1Tflops .

 

[Sound_Card]

why do people say that G80 has 520gflops when it has a missing mul and is truly rated at 3xxgflops?

I think R600 has alot more flop power than Dave O. would like us to believe. I'm thinking 700gflops.

 

[satein]

The subject line says it all: Who will be the first GPU maker to ship a single chip GPU solution with over 1TFLOPs of programmable performance? NV, ATI/AMD, Intel, or other?

And a bonus speculation question: WHEN?

I believe if they are not too busy on the Fusion stuff, it could be achieved in Q1/Q2 2008. 65nm process would be possible on packing all that 1T Flops onto a single GPU core. But if not, and it needs 45nm process, the timeframe can be shifted to Q4/2008...

Hopefully, your question would dedicate to only a single GPU Flop count not multiple GPU Flop count.

 

[Anarchist4000]

I'm going with ATI on the refresh of R600. If they move to 65nm and keep the chip size about constant instead of reaping the savings on die size it will likely have >1TFLOP. Didn't Orton actually say maybe in the future over 1TFLOP or something along those lines?

I'm still guessing R600 will be ~768GFLOPs.

 

[Razor1]

he said in excess of a half a teraflop, and I think AMD will be first to get to the teraflop level as well, but again depends on execution on thier side. Within a year I would suspect.

 

[Ateo]

Flops are useless...
This was what Anandtech ahd to say in a pulled console article:

Another way to look at this comparison of flops is to look at integer add latencies on the Pentium 4 vs. the Athlon 64. The Pentium 4 has two double pumped ALUs, each capable of performing two add operations per clock, that's a total of 4 add operations per clock; so we could say that a 3.8GHz Pentium 4 can perform 15.2 billion operations per second. The Athlon 64 has three ALUs each capable of executing an add every clock; so a 2.8GHz Athlon 64 can perform 8.4 billion operations per second. By this silly console marketing logic, the Pentium 4 would be almost twice as fast as the Athlon 64, and a multi-core Pentium 4 would be faster than a multi-core Athlon 64. Any AnandTech reader should know that's hardly the case. No code is composed entirely of add instructions, and even if it were, eventually the Pentium 4 and Athlon 64 will have to go out to main memory for data, and when they do, the Athlon 64 has a much lower latency access to memory than the P4. In the end, despite what these horribly concocted numbers may lead you to believe, they say absolutely nothing about performance. The exact same situation exists with the CPUs of the next-generation consoles; don't fall for it.

The same goes for GPU's Flops are useless...
Thanks to Sony and Microsoft for getting Flops PR-spinned

 

[jamis]

Easy, ATI/AMD R6x0 65nm by the end of the year.

Edit: Well, provided the company is still up and running.