Intel integrated GPU

[kriesel]

Originally Posted by masser

I have the latest mfakto executable on the laptop, but the self-tests are failing 14 out of 30, of the short test set. Any ideas what I should try?

Look for patterns to the fails. If they're all above 80 bits, it could perhaps be used in the 70s.

[M344587487]

I got an old intel iGPU, HD530, to run OpenCL on Ubuntu 20.04 simply by installing these packages from the standard repos:

sudo apt install intel-opencl-icd opencl-headers clinfo ocl-icd-libopencl1

AFAIK intel-opencl-icd is the intel ICD, ocl-icd-libopencl1 is the generic ICD loader for OpenCL, opencl-headers gives you the headers for compiling and clinfo is just the utility to check the status of OpenCL devices. There may be a meta package that sets up a working intel OpenCL environment for you containing these packages and maybe more but I didn't come across it.

gpuowl compiled and ran as-is, didn't even have to add -L/usr/lib/x86_64-linux-gnu to LIBPATH as I guess it's in a standard location.

[PhilF]

Originally Posted by M344587487

I got an old intel iGPU, HD530, to run OpenCL on Ubuntu 20.04 simply by installing these packages from the standard repos:

sudo apt install intel-opencl-icd opencl-headers clinfo ocl-icd-libopencl1

AFAIK intel-opencl-icd is the intel ICD, ocl-icd-libopencl1 is the generic ICD loader for OpenCL, opencl-headers gives you the headers for compiling and clinfo is just the utility to check the status of OpenCL devices. There may be a meta package that sets up a working intel OpenCL environment for you containing these packages and maybe more but I didn't come across it.

gpuowl compiled and ran as-is, didn't even have to add -L/usr/lib/x86_64-linux-gnu to LIBPATH as I guess it's in a standard location.

Great timing! I was just looking into how to install OpenCL for an Intel GPU and found your post, which I was able to use to get OpenCL running on Debian by simply substituting "beignet-opencl-icd" in place of "intel-opencl-icd". The rest of the packages installed normally.

I haven't actually computed with it yet, but clinfo's output looks good.

Thanks!