Katherine Chen
Katherine Chen

Reputation: 542

Found no NVIDIA driver on your system error on WSL2 conda environment with Python 3.7

I have a Nvidia 1080Ti GPU, and I want to run Pytorch on WSL2, but I got error "Found no NVIDIA driver on your system" but I did installed the NVIDIA driver. Here is the step I did.

  1. I installed WSL2, and installed NVIDIA driver for Cuda on WSL from GeForce Driver: https://developer.nvidia.com/cuda/wsl/download

  2. I activate a clean conda environment with Python 3.7

  3. Then I run the Pytorch installation: conda install pytorch torchvision cudatoolkit=10.2 -c pytorch

Then the error occurred saying Found no NVIDIA driver. I came across a post on Pytorch forum, and someone did get it to run in a similar settings: Ubuntu 18.04 + Conda + Pytorch https://discuss.pytorch.org/t/found-no-nvidia-driver-on-your-system-but-its-there/35063/4

I don't have multiple GPUs, so I don't know how to get my driver recognized in WSL2. Thank you for any thoughts!

Upvotes: 7

Views: 14739

Answers (2)

Katherine Chen
Katherine Chen

Reputation: 542

@onoma was right that the original installation step was missing one part, which is the Windows Insider build.

Found this document has the full steps guide for installing WSL2 with Cuda support. In case anyone who come across this post would be interested:

https://docs.nvidia.com/cuda/wsl-user-guide/index.html

Note: Education version doesn't seem to have option to opt-in Windows Insider Program. On my system, I can access Windows Insider Program in the Windows Settings, but it requires the Diagnostic data set to full to get the build. However, no such option in the settings that I can configure on, and it shows "Some of the settings are hidden or managed by your organization".

Upvotes: 5

onoma
onoma

Reputation: 1465

What's your Windows version? (Run winver.exe) You need to run a Windows Insider build 20145 or superior in order to use CUDA in WSL2.

You will know the gpu is detected if /dev/dxg file exists.

Upvotes: 8

Related Questions