OpenCL ray tracing benchmark tested with Intel OpenCL SDK Alpha

[jogshy]

Hi,

I've just tested your new Intel OpenCL SDK with ratGPU ( www.ratgpu.com ) which is an OpenCL ray tracing renderer / benchmark and your implementation seem to work well:

[ARNON_P_Intel]

Great,

We will be glad to learn more from your experience as you are moving forward with your experiments.

[jogshy]

Tested it with your new Alpha SDK x64:



Your SDK x64 gives a nice speed boost ( 503s vs 565s ) plus I've found it very stable. Good job!

[jogshy]

Test with OpenSUSE 11.4 x64 and your Intel OpenCL 1.1 SDK beta for linux:




Works like a charm! Good job!
Btw, I think you should provide your CL sdk using a .deb / tar package for other distros.... and a x86 version won't neither hurt :p

[ARNON_P_Intel]

Thanks for the feedback, we do appreciate this.

Great to see this applications running on all configurations.

Have you managed to use our tips and tricks and optimize the application respectively?

[jogshy]

The program currently has 3 optimization paths: one for NVIDIA, other for ATI and other for Intel ( although I'm waiting for your final SDK and GPU support to make heavy optimizations ).

I try to follow your tips and tricks guide, yep. However I have a doubt. What's better, a direct memory access to a float4 pointer or a vload? Will be fantastic if you put in that guide some advices about that, pls.