Test Nvidia, AMD GPUS and CPUs (CUDA, CAL, CPUs)
Key points of benchmark:
*Uses very high performance codes:
*Win x32,x64! (big win in x64 in multicore CPUs)
*Single precision and double precision!
*Test any size!
Download!
Also Works on Wine! (at least it worked with the 32 binaries with Wine Wrappers around CUDA and CAL I did a year ago but which I think I sadly lost! anyway where based on ideas around CUDA Wine wrapper for Folding@home)
There were issues in CPU using only one thread altough I forced using it via OpenMP env variables and Intel MKL env variables..
Also I tested x64 with a CAL Wine64 wrapper.. but crashed..
Uses very high performance codes:
*For Nvidia GPus uses CUBLAS fast matmul (volkovs code)
*CAL matmul uses AMD IL from AMD SDK
(anyway better is known see beyon3d 1tflop matmul)
*Uses Intel MKL libs (hopefully also using new dll's you get AVX acceleration (at least I hope so..))..
Also assuming Larrabe executes normal executables and has a intel MKL library with binary compatibilty then
also that..
I done it past year!
for testing a 8800GT and 4850!
now with Windows 7 you can use to test simultaneuos perf on ATI and Nvidia and CPUs!
Still working only need to use HxD to change references from amdcal to atical DLL's
There I can remove the dll's as a part of AMD driver and I get also new GPU support automatically..
tested on ATI 5850..
Source code: sorry guys but I would have to search very hard for finding it! it's old cde..
Results:
Core i7 920
ATI 5850
GTX 275
Build info: X86 Release
Build date: Jul 31 2008 20:45:28
Machine arch: little endian.
Number of CPU cores: 8
Timer used: MULTIMEDIA TIMER
Freq. of TIMER: : 2.63554 Mhz.
CPU: 2716.93 Mhz.
Size: 4096
CAL tests:
=========
Float
Tam: 4096 Time running: 0.369012 s. Gflops: 744.902950 Gflops.
Testing CAL matmul double
Double
Tam: 2992 Time running: 0.483039 s. Gflops: 221.800549 Gflops.
CUDA tests:
==========
There is 1 device supporting CUDA
GPU 1: "GeForce GTX 275" SM:1.3
MP: 30 Cores: 240 Freq:1.40 GHz
Mem: 896Mbytes BW (Pinned) H2D: 2.65GB/s D2H:2.81GB/s D2D:55.3
CUDA floating code
Time running: 0.432000 s. Gflops: 318.106889 Gflops.
CUDA no IO floating code
Time running: 0.339000 s. Gflops: 405.375151 Gflops.
CUDA double code
Time running: 1.844000 s. Gflops: 74.523957 Gflops.
CUDA no IO floating code
Time running: 1.682000 s. Gflops: 81.701651 Gflops.
CPU tests:
==========
Intel MKL code
Time running: 2.573000 s. Gflops: 53.409318 Gflops.
Double
Time running: 5.825000 s. Gflops: 23.591790 Gflops.
OK.
Sunday, 8 November 2009
Subscribe to:
Post Comments (Atom)
0 comments:
Post a Comment