[iotku]

The Nvidia container toolkit works pretty much out of the box on WSL these days as well.

Funny that some cuda stuff works better through Windows virtualizing Linux than Windows natively, but if we're being honest even as a native Linux user, WSL probably provides a better user experience (vs having to use Nvidia drivers on Linux anyways)

[theturtle32]

Having had the displeasure of trying to get CUDA running under WSL2, I can tell you it is most definitely not a better user experience :-P

[nightski]

Weird it was pretty much turn key here. Maybe that was the early days when it was in beta?

[freeone3000]

Is it a better user experience than trying to get CUDA natively? :P

[angm128]

My personal experience: Fedora 38- just needed to compile gcc-12 (took very long), Windows- Installer failed

This clearly shows that GNU/Linux must be superior

[vunderba]

Is the container toolkit even necessary anymore in WSL2? I did a fresh install of windows 10 LTSC a few weeks ago and installed WSL2 and docker with WSL2 integration and I was able to use my nvidia rtx card (checked through nvidia-smi) without any issues.

[anvuong]

It registered the card, as in you can query the device id using nvidia-smi. But CUDA/cuDNN requires further work.

[vunderba]

Interesting. I'll have to check to be sure, but I think maybe something is happening automagically if you have reasonably up to date nvidia drivers on the host OS, because I was able to run the EmotiVoice TTS docker (which requires nvidia gpu) from WSL2.

https://github.com/netease-youdao/EmotiVoice

[holoduke]

What? No latest windows drivers give full cuda support in wsl out of the box. I am running all kinds of models without any issues at all.

[idonotknowwhy]

Hasn't been my experience. Windows keeps a few hundred mb of vram tied up for the gui, Linux only holds 14mb per GPU

[ekianjo]

Nvidia drivers work just fine on Linux... unless you are clueless at following instructions

[iotku]

Counterpoint:

https://forums.developer.nvidia.com/t/545-drivers-have-bad-f...

https://forums.developer.nvidia.com/t/wayland-native-wayland...

https://gitlab.freedesktop.org/xorg/xserver/-/issues/1317

and literally me being a Linux Nvidia 1080ti user for years and having plenty of issues

https://i.imgur.com/lwHDZoq.png

[ekianjo]

I thought we were talking about CUDA here?

[nsagent]

As someone who has had a home ML server since 2016 with two TitanX GPUs, and has worked on and maintained numerous servers since then I can definitely echo that maintaining Nvidia drivers, along with CUDA, CUDNN, etc has always been a hassle. It's certainly gotten better over time, but it's still quite fragile.

Automatic kernel update? https://forums.developer.nvidia.com/t/nvidia-smi-not-working...

What about upgrading to a new version of CUDA? https://stackoverflow.com/questions/43022843/nvidia-nvml-dri...

What about trying something like enabling forward compatibility for CUDA using an older driver? https://discuss.pytorch.org/t/torch-is-unable-to-detect-cuda... (This issue was actually just posted within the last day, so clearly people still have problems.)

If you haven't run into any issues, then I'd say you're very lucky. Just don't pretend lots of others haven't run into issues.