Who will deliver the first 1TFLOPs retail GPU?
- Thread starter Acert93
- Start date
Anarchist4000
Veteran
Razor1 said:
I thought it was R600 would be in excess of half a teraflop and in the future possibly around a full teraflop. Hence R600 is between 500GFLOPs and 1000GFLOPs and the refresh possibly breaks the barrier.
Ateo said:
Except the way GPUs are set up they can actually feed the ALUs so you get much closer to the theoretical numbers. Plus with GPUs almost everything is a float so the numbers at least give a good idea of the capability. FLOPs are cheap so you never want them to be holding down performance.
icecold1983
Banned
ati in 2nd quarter of 08
edit - actually on further consideration im saying q1/q2 09. dont rly see a gpu 3x as fast as g80 coming until at least early 09
edit - actually on further consideration im saying q1/q2 09. dont rly see a gpu 3x as fast as g80 coming until at least early 09
Last edited by a moderator:
Flops are useless...
This was what Anandtech ahd to say in a pulled console article:
The same goes for GPU's Flops are useless...
Thanks to Sony and Microsoft for getting Flops PR-spinned
Hehe at first it was bits, then MHz, MIPS, BOPS, FLOPS. What's next qubits?
Hehe at first it was bits, then MHz, MIPS, BOPS, FLOPS. What's next qubits?
I would settle for actual preformance/IQ
It's not how many bits get altered it's who the end result looks and preforms that matters.
FLOPS tells you nothing at all...
They tell you quite a lot for GPGPU work, assuming the efficiency is decent, actually...
Anarchist4000
Veteran
Ateo said:
There are certain situations in graphics and especially GPGPU where FLOPs are very important. In the past they haven't always made that big of a deal because texturing was typically holding you back. As shaders get bigger there is less texturing so you're more likely to use all that processing power as the bottleneck shifts.
This next generation could be huge for the GPGPU and physics communities so it's not hard to imagine using all the FLOPs. With CUDA and CTM there are a whole lot of really interesting applications for video cards that haven't been seen before. I'm sure once someone makes a CUDA/CTM backend for something like ffdshow people will be glad those FLOPs are there.
They tell you quite a lot for GPGPU work, assuming the efficiency is decent, actually...
Nice "rubber-phrase"
Mintmaster
Veteran
I would have voted for ATI because they seem to be focussing on math more than NVidia, but I don't think ATI will hit 1 TFLOP with R600, and I think NVidia will get their next gen out first.
Gotta go with NVidia here.
Gotta go with NVidia here.
Who said anything about beeing 3X faster than G80, this was about them damn FLOPs. So I figure R600 is going to be twice the FLOPs of R580, then later on this year R6x0 with around 25% more shaders and higher clocks and there you'll have it.
SF = 1/4 FLOP per cycle.
It's 3,25 FLOP per cycle per "stream processor".
But: Who will deliever the first 500 GFLOP/s GPU?
Similar threads
