MKL on NVIDIA GPU

[marcsolal]

If I am posting at the wrong place, please let me know where I should post this. 

I have a code which is using MKL a lot and I am now looking to use this on a gpu. For this reason, I am very interested by the new oneMKL. I am expecting to be able to go to GPU easier. Simply, even after the webinar, it is not clear to me when it will be possible to use OneMKL on NVIDIA GPU. I understood somebody is developing a OneAPI compiler compiler which is compatible to NVIDIA GPUs. I did not undertand who is doing this, what is the status, and if this would be enough to use MKL on NVIDIA GPUs or if something else is needed. Could you please clarify? I need to decide if I need to go to CUDA and CUDA librairies or if I can stay with MKL, which I would prefer.

Thanks.

[Ruqiu_C_Intel]

Hello marcsolal,

Thanks for your questions.

Intel oneMKL can run on both Intel CPU and Intel GPU.

For oneMKL runs on NVIDIA GPU, there is third party (codeplay) working on it. Please check codeplay website for more information.

And, Intel has developed compatibility tool to convert CUDA libraries to DPC++, after compile it with DPC++ compiler, then it can run on Intel GPU.

Best Regards,

Ruqiu