NVIDIA OpenCL Linux Benchmarks

[evolution]

Do you even know what you're talking about? GT2xx is NOT Fermi.

Yes I know that GFs GT2xx aren't Fermi, only GFs4xx...

Cheers

[deanjo]

The only leap in GPGPU performance since 8000 series was the GTX480/470/465. GTX460 had its GPGPU capability scaled back. And the 500 series apparently got the enhancements back. But when you're looking at 9000 vs 200 series, it's the same architecture. I think only the video engine was changes, besides the number of pipes.

Not even close, many changes occurred in the computing capability in the GTX 2xx series (64-bit precision being one big one). In fact many generations have carried changes.

You might want to look at this for a good comparison:

http://www.geeks3d.com/20100606/gpu-computing-nvidia-cuda-compute-capability-comparative-table/

[deanjo]

Just to give an idea of what the GTX-580 does here are some results. There is no smallPT results as there is a bug in smallpt tests that causes it to report negative values when the values are too large to fit into its integer type according to Micheal.

[deanjo]

I should note that the gluxMark is not directly comparable either. Micheal ran them @ 1920x1080 vs my 1920x1200.

[deanjo]

But how about actually comparing it to same benchmarks and hardware but using MS Windows? This will allow real comparission.

From the benches I've ran they are pretty much both even in performance (small enough deviations to be deemed irrelevant) , at least with nvidia hardware.

[skeetre]

Can anyone tell me how to get the OpenCL stuff to run? I've installed the latest Nvidia drivers, 260.19.29, but when I install the phoronix-test-suite packages it says I need to install the Nvidia OpenCL package. I've googled and looked on the forums, but I can't find anything about how to get OpenCL to work.

I have an 8800GT, gtx 460, and a gtx 470 that I'd like to benchmark.

[deanjo]

The benches require that the openCL headers be placed in /usr/include/CL/. The headers you can get when you install the cuda sdk available here:

http://developer.nvidia.com/object/cuda_3_2_downloads.html

[skeetre]

Thanks! I'll give that a shot.

[bug77]

Not even close, many changes occurred in the computing capability in the GTX 2xx series (64-bit precision being one big one). In fact many generations have carried changes.

You might want to look at this for a good comparison:

http://www.geeks3d.com/20100606/gpu-...arative-table/

I don't see anything about 64bit precision over there. I see more registers, more threads and warps per multiprocessor and couple of atomic operations support. I'm not really familiar with gpu computing, but these all look like minor improvements to me.

[deanjo]

I don't see anything about 64bit precision over there. I see more registers, more threads and warps per multiprocessor and couple of atomic operations support. I'm not really familiar with gpu computing, but these all look like minor improvements to me.

Let me highlight for you:

http://en.wikipedia.org/wiki/Double_precision_floating-point_format