Intel_extension_for_pytorch on NVIDIA GPU

[rapidRabbit]

Hello

I am aware, that intel-provided implementation of oneAPI AI Toolkit does not support NVIDIA GPU.

I was trying to use optimization tools provided by particular part of AI Toolkit - intel_extension_for_pytorch on NVIDIA GPU. 

I used the plugin provided by Codeplay  - oneAPI for NVIDIA GPU.

The problem is, that torch version built for GPU usage along with intel_extension_for_pytorch does not see GPU hardware (torch.cuda.is_available() == False), while sycl-ls sees the GPU:

```

[ext_oneapi_cuda:gpu:0] NVIDIA CUDA BACKEND, NVIDIA GeForce RTX 3070 8.8 [CUDA 12.1]

```

The driver/cuda is also installed.

I am curious whether the usage of intel_extension_for_pytorch on NVIDIA GPU possible at all?

If no, if any part of Intel oneAPI AI Toolkit providing enhancement for pytorch/tensorflow can be used on NVIDIA GPU?

[Rahila_T_Intel]

Hi,

Thanks for posting in Intel Communities. 

We are looking on your issue internally.

Thanks

[rapidRabbit]

Hi,

do you have any update for the moment?

Thanks 

[YuningQiu_Intel · Accepted solution]

Hello,

Thanks for raising this question. This is valuable to us. However, currently we have not verified if IPEX supports NVIDIA cards, and the plan to verify the usability is not decided yet. 

Currently, CodePlay is supporting oneAPI Base Toolkit, but the support on oneAPI AI Toolkit is still WIP.

Thanks again and please let me know if you have any further questions.