Re: Package request: cuda

[bobertlo <bobertlo@users.noreply.github.com>]

[-- Attachment #1: Type: text/plain, Size: 1318 bytes --]

New comment by bobertlo on void-packages repository

https://github.com/void-linux/void-packages/issues/31984#issuecomment-1002598948

Comment:
To my surprise, CUDA is included with the proprietary NVIDIA packages.

```
nvidia-libs-470.94_1	/usr/lib/libcuda.so -> /usr/lib/libcuda.so.470.94
nvidia-libs-470.94_1	/usr/lib/libcuda.so.1 -> /usr/lib/libcuda.so.470.94
nvidia-libs-470.94_1	/usr/lib/libcuda.so.470.94
```

I was able to install pytorch, `torch.cuda.is_available()` returns true, and the mnist example trains a model.

This issue could be closed (edit maybe not?) but I view this as a documentation issue since there is absolutely no mention of cuda in any package descriptions or documentation. If you google cuda void linux you find multiple open package requests and reports that cuda is not available or supported.

EDIT: I am unclear about what else is generally provided with CUDA. I think conda is actually pulling in a lot of the runtime. I will try to look into what is missing and may be required for other native packages wanting cuda. What is included already is enough for me, and I think enough to get nvidia-docker running. I wanted to share this because I was about to just multiboot (while looking at possibly packaging cuda) based on the information I found about cuda on void.