• Competitor rules

    Please remember that any mention of competitors, hinting at competitors or offering to provide details of competitors will result in an account suspension. The full rules can be found under the 'Terms and Rules' link in the bottom right corner of your screen. Just don't mention competitors in any way, shape or form and you'll be OK.

DirectCompute benchmark

Rroff said:
We don't know that...
Click to expand...

The 5700 series are scoring almost the same as the 5800 series despite having a lot less shaders and memory bandwidth, how can you say there isn't something wrong with the benchmark? ATI's drivers can't be that broken in Direct Compute.
 
Broken Hope said:
The 5700 series are scoring almost the same as the 5800 series despite having a lot less shaders and memory bandwidth, how can you say there isn't something wrong with the benchmark? ATI's drivers can't be that broken in Direct Compute.
Click to expand...
but maybe some of the 57x0 are overclocked so maybe that why they are close.
 
hyperst said:
system requirements state dx11

heres mine, not bad for 5770 owning an above 5870:D

Capture-5.jpg
Click to expand...
overclocked? also crossfire or single?
 
Theres something not entirely correct with it possibly - I'm getting an almost identical score here clock for clock testing with a 260GTX SLI against 295GTX despite the 260GTX SLI having 48 less SPs.
 
Rroff said:
Theres something not entirely correct with it possibly - I'm getting an almost identical score here clock for clock testing with a 260GTX SLI against 295GTX despite the 260GTX SLI having 48 less SPs.
Click to expand...

Yeah, there's also very little separation between the 5700 and 5800 series cards, too. It sounds almost like the benchmark is more limited by the card's clock speed than its computational resources. That would likely explain the GTX200 series' unrealistically large advantage, at any rate (huge shader clock speed). I think it's possible the workload isn't large enough to really take advantage of the wider graphics chips' computational resources. If anyone could test on a 9800 GTX+ or something similar that'd really clear the issue up.
 
Looks like to me the benchmark isn't using all of the RV870's superscaler "stream processors", it's only using the first SP unit instead of all 5.

While the nVidia chip is scalar and thus all the SP's are being used.

Still a bit poor from AMD/ATI.
 
Rroff said:
Theres something not entirely correct with it possibly - I'm getting an almost identical score here clock for clock testing with a 260GTX SLI against 295GTX despite the 260GTX SLI having 48 less SPs.
Click to expand...

could pci-e lane speed be different? an i7 vs a core2quad is massive difference in this benchmark so if the gtx295 got less cpu power maybe it makes a difference?


guess this benchmark shows the difference between ati's shaders vs nvidia's cuda cores (stream processors). nothing new though, the gpgpu performance of nvidia cards is astounding, im shocked at how slow the ati 5 series cards are in this test.
 
Mr Paul said:
As I said Cyber, I'm sure all the SP's are not being used. Only 1 per block of 5.
Click to expand...

guess only time will tell. but if this benchmark uses pure stream processing then ati's figure of 1600shaders is useless since its really 1600/5 = 320 "stream processors" which in turn is pathetic in performance if its getting beat by an old skool gtx260.

be interesting to see if ati pull out some sort of driver to boost performance of if its true that the new 5 series cards are just for gaming and not computing.


im expecting fermi to be setting some kind of records in this test since fermi is all about compute power.
 
Cyber-Mav said:
could pci-e lane speed be different? an i7 vs a core2quad is massive difference in this benchmark so if the gtx295 got less cpu power maybe it makes a difference?


guess this benchmark shows the difference between ati's shaders vs nvidia's cuda cores (stream processors). nothing new though, the gpgpu performance of nvidia cards is astounding, im shocked at how slow the ati 5 series cards are in this test.
Click to expand...

I'm comparing it against a 295GTX here thats on a core2quad, only differences are the 260GTX SLI is on a board using the NF200 chip (but I think the 295GTX has one onboard anyhow) the 295 is on a slightly slower CPU but I wouldn't have thought it made that much difference.
 
Rroff said:
I'm comparing it against a 295GTX here thats on a core2quad, only differences are the 260GTX SLI is on a board using the NF200 chip (but I think the 295GTX has one onboard anyhow) the 295 is on a slightly slower CPU but I wouldn't have thought it made that much difference.
Click to expand...

clock your cpu down a little and then try the benchmark. all you got to do is go into bios and drop the multi a notch and run the bench.
 
Cyber-Mav said:
guess only time will tell. but if this benchmark uses pure stream processing then ati's figure of 1600shaders is useless since its really 1600/5 = 320 "stream processors" which in turn is pathetic in performance if its getting beat by an old skool gtx260.

be interesting to see if ati pull out some sort of driver to boost performance of if its true that the new 5 series cards are just for gaming and not computing.
Click to expand...

A lot of this has to do with how well the programmer has vectorised their code, too (to anyone who doesn't understand that, basically, this means how well they've managed to group together like operations, say for example lots of multiplications need to be done, in code). This isn't just an R700-architecture optimisation, either, as CPU's that support SIMD instructions (SSE, 3DNow!, MMX, etc.) also need vectorised code in order to take advantage of those units.

im expecting fermi to be setting some kind of records in this test since fermi is all about compute power.
Click to expand...

I have no doubts it will, given what Nvidia have given us about that it looks like it should be an utter monster in the GPGPU arena. :D
 
I think the reason NV cards are getting high scores over ATI, is to do with NV being much better at GPU computing then ATI.. taking Folding for example, I may well be wrong.
 
Check the forums out for this bench, its a POS, GTX 280's scoring 30000, so I would say forget this bench as it looks poorly coded.
 
ok guys the benchmark hasnt been coded with ati hardware in mind , read it on tech powerup forum

ati guys please rerun the bench with the single thread ticked , and nvidia guys do the same that way we are all useing one bank of shaders , see what the scores are then

Capture-7.png
 
The single threaded score should be identical to the multi threaded score when it comes to the GPU score.

"S" for CPU means singlethreaded code while "S" for GPU means passing D3D11_CREATE_DEVICE_SINGLETHREADED flag to the DirectX11 device (could be misunderstood as a singlethreaded code execution on GPU)
Click to expand...
 
You must log in or register to reply here.
Back
Top Bottom