How did arrive at that conclusion?
750 x 800 x 2 = 1.2 TFlop
edit - You might have done 750 x 480 x 2 = 720 GFlops
AMD: R7xx Speculation
- Thread starter Unknown Soldier
- Start date
- Status
750 x 800 x 2 = 1.2 TFlop
edit - You might have done 750 x 480 x 2 = 720 GFlops
Silent_Buddha
Legend
For GPGPU: ~1TFLOP in ATI form costs $200 and in NVidia form costs $600.
Comparing double-precision: 300+ versus ~125 GFLOPs.
For $600 you can have 1 TFLOP of ATI's double-precision or 125 GFLOPs of NVidias
Hmm.
Jawed
Which makes it a bit of a shame that ATI didn't also release a high level programming language for GPGPU ala CUDA. But instead focused on a more low level approach to programmability.
It appears (from my limited perspective) that GPGPU is taking off on Nvidia hardware mostly due to the "relative" ease of use of CUDA.
Still, at least it's seen a bit of a resurgence in the professional market for FireGL. I can only imagine FireGL based on Rv770 is going to be even more popular.
And for GPGPU, it's going to be hard for data centers to ignore the power consumption difference between Rv770 and GT200.
Regards,
SB
AMD Stream Computing
CTM : Close To Metal (Low-level language)
CAL : Compute Abstraction Layer
Brook+ (High-level language / ANSI C / BrookGPU adaptation)
ACML: AMD Core Math Library
APL: AMD Performance Library
COBRA: Video library for video transcoding acceleration
Last edited by a moderator:
You want to download the following file:
http://rapidshare.com/files/120645188/R700.jpg | 193 KB
The download session has expired.
EDIT: Odd, I had to click through a few screens that kept giving the same error, then it started working. The one which worked for me.
Are you talking about the legendary ATi chip, which will be revealed in approximately 6 months ?
lol: I think to get basic idea how (2x RV770) Radeon 4870X2 will fight against huge GT200.
Well, G80 had less than R600, but then reality set in. Same for G92 vs RV670.
Remember that theoretical teraflops numbers don't mean much unless there an actual advantage in real-world applications.
Still, let's wait and see what RV770 can do. It sure looks promising.
First I would like to post all of Ailuros' posts one after one, it might help someone decipher the code...
The only thing I can think of that is relatively "creative" would be 80 5d equaling 400 shaders each doing 4Flops at 750mhz means 1.2TFlop.
80/32 = 2.5:1
80/24 = 3.3:1
4 or 5 seems common, so let's go out of the box and say 6.
5 FLOPs/ALU in both cases. How about a bit more creative math?
The only thing I can think of that is relatively "creative" would be 80 5d equaling 400 shaders each doing 4Flops at 750mhz means 1.2TFlop.
80/32 = 2.5:1
80/24 = 3.3:1
Last edited by a moderator:
19 days to go... and we are still in the dark
lol, good work!
I'm going to play devil's advocate here and go with the simple, boring and realistic:
750Mhz core
96 5D processors
720 Gflops
32 TMUs
Should be more than enough for +50% or more on RV670.
Speculating that we will get a 150% increase in shaders and 100% increase in TMUs for a 25% increase in transistors is too far beyond the realm of common sense for me. That would mean RV670 was made mostly of vanilla pudding and they decided to swap it for transistors this time around.
750Mhz core
96 5D processors
720 Gflops
32 TMUs
Should be more than enough for +50% or more on RV670.
Speculating that we will get a 150% increase in shaders and 100% increase in TMUs for a 25% increase in transistors is too far beyond the realm of common sense for me. That would mean RV670 was made mostly of vanilla pudding and they decided to swap it for transistors this time around.
That would work, other than your math of doing 2Flops per shader but putting 3Flops in definitely makes it better.
96*5*3*750=1.08TFlop
But that goes against the TFlop target.
Maybe they got rid of something?
I think key is to work with 4850's clock (625MHz) to reach the TFlop mark ..
- Status
Similar threads
