cuda devices not detected by DPC++ and sycl::device::get_devices

[Daniel_D]

Hi,

I try to convert some of my cuda kernels to sycl. I'm on windows 11 and VS2023 latest updates installed. I cvan use my cuda device with my cuda c++ kernel without any issues. It is detected from the OS.

Unfortunately none of them arte detected by sycl and I get only the UHD GPU and the Intel CPU from a call to sycl::device::get_devices()

Any idea what I'm doing wrong? Maybe I need an additional driver to make cuda devices available to sycl?

Thanks,

Daniel

[NoorjahanSk_Intel]

Hi,

Thanks for posting in Intel Communities

Intel oneAPI DPC++ compiler can be used only with Intel GEN9 and higher GPUs.

Please find the below link to get details on Intel® oneAPI DPC++/C++ Compiler System Requirements:

https://www.intel.com/content/www/us/en/developer/articles/system-requirements/intel-oneapi-dpcpp-system-requirements.html

If you want to use the DPC++ compiler with NVIDIA GPU you can use the open-source DPC++ compiler.

Please refer to the below link regarding the opensource DPC++ compiler:

https://intel.github.io/llvm-docs/GetStartedGuide.html

These forums are intended to support queries related to Intel Products. Here, we do not provide support to the non-Intel GPU. 

We suggest you raise a GitHub issue in the below link to get support for non-Intel GPUs

Please find the below GitHub link to raise a query: https://github.com/intel/llvm/issues

So, could you please confirm whether we can go ahead and close this thread from our end?

Thanks & Regards,

Noorjahan.

[Daniel_D]

Hi Noorjahan,

thanks for your fast reply. I was under the wrong impression that nvidia/amd devices were supported by your dpc compiler. When I read: Migrate CUDA* to DPC++ Code: Intel® DPC++ Compatibility Tool I didn't notice that you do not support these devices und use the tool just to get cuda code running on the intel CPUs/GPUs.

Will the open source DPC compiler for cuda also support the intel CPUs/GPUs or do I need one DPC compiler for each target i want to use (Intel CPUs, Intel GPUs, AMD GPUs and nVidia GPUs)?

Yes, you can close this issue.

Thanks,

Daniel

[NoorjahanSk_Intel]

Hi,

>>Will the open source DPC compiler for cuda also support the intel CPUs/GPUs

Yes, you can target Intel CPUs/GPUs along with AMD GPUs and NVIDIA GPUs using the open-source DPC++ compiler.

Intel oneAPI DPC++ compiler supports only Intel CPUs/Intel GEN9 and higher GPUs.

We request you to raise an issue in GitHub for further queries on the opensource DPC++ compiler

https://github.com/intel/llvm/issues

>>Yes, you can close this issue.

Thanks for the confirmation. If you need any additional information regarding Intel products, please post a new question as this thread will no longer be monitored by Intel.

Thanks & Regards,

Noorjahan.

[encalle]

What DPC++ version are you using? By analyzing the .dll, I discovered that CUDA PI plugin requires in the newer DPC++ versions, the file cupti.dll in windows at least, and that file is only available with the CUDA Computing toolkit. If the plugin interface don't find it, it will fail silently. I fixed it adding:

"C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v11.8\extras\CUPTI\lib64" directory to my %PATH%

[Daniel_D]

Hi,

I was using the latest available package from Intel. But I always run into new problems that stoppend development. I changed to OpenCL. I will look into DPC++ in the future.

Thanks, for your reply.