Archives Discussions

[kacperpl1]

Opteron OpenCL Support

I'm new to this forum so hello guys and hope we'll get along

I'm using JOCL library to compute 13 500 000 000 000 combinations of kernel with 5 parameters and I'd like to use full power of my box, however there's a problem with support for my CPU's.

Requirements of SDK by AMD for cpu:

"Supported Processors: X86 CPU w/ SSE 2.x or later"

Workstation #1:

2 x Opteron 270 ( Supports SSE 1, 2 & 3 )

Asus K8N-DL

Geforce 9600GT

Windows 7 32 bit

Stream SDK 2.2 ( most recent)

GPU caps says my cpu doesn't support OpenCL and I get CL_DEVICE_NOT_FOUND when trying to run my app on CPU only

(notice that GPU and ALL modes works)

What might I do wrong that CPU isnt supported?

[himanshu_gautam]

Are you able to run any samples?If so Please post the clinfo output.

Otherwise you should try reinstalling the SDK.

[kacperpl1]

I'm unable to run opencl benchmarks for cpu on gpu caps and I'm unable to run my program in cpu context mode.

[kacperpl1]

I've found the direct cause:

If you install nvidia CUDA sdk and after that you install Stream SDK you get no CPU support. Its not the problem with my CPU's but with CUDA sdk's OpenCL not supporting CPU. Because of this there are two problems to be solved by Stream SDK creators:

Make CPU support when CUDA SDK installed

Roll back with simply uninstalling CUDA sdk - ive done that, however the CPU support didn't come back, even after reinstalling Stream sdk - ive had to completely rollback my system to time before cuda drivers and sdk install.

Hope you'll they'll make it in next release

[himanshu_gautam]

Can you provide which CUDA SDK you were installing.

[kacperpl1]

Most recent at that time which was 3.2. I have reproduced the bug on Intel C2QE and Quadro FX1700 with the same sdk and the only thing to roll it back to work for CPU only was to roll back system to the save point before installing the cuda sdk and then install only the stream sdk.