Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

Add Windows WSL2 instructions #44

Merged
merged 2 commits into from
Nov 20, 2022
Merged

Add Windows WSL2 instructions #44

merged 2 commits into from
Nov 20, 2022

Conversation

randyau
Copy link
Contributor

@randyau randyau commented Jun 11, 2022

Notes for compiling jax w/ CUDA support in Windows WSL environment

@joeltgray
Copy link

Any chance of getting one for AMD GPUs using ROCm and Windows WSL?

@Phildo
Copy link

Phildo commented Jun 15, 2022

followed instructions (including installing jax from source, which takes ~1hr! would recommend adding a warning about that), and am still getting the "warning": WARNING:absl:No GPU/TPU found, falling back to CPU. (Set TF_CPP_MIN_LOG_LEVEL=0 and rerun for more info.), with the frontend failing to connect to the server.

Any ideas what I could be missing?

Also a note: I think jax fixed the "config error" you warned about and provided a workaround for, so I don't think that's necessary any longer.

@matthew-lowe
Copy link

followed instructions (including installing jax from source, which takes ~1hr! would recommend adding a warning about that), and am still getting the "warning": WARNING:absl:No GPU/TPU found, falling back to CPU. (Set TF_CPP_MIN_LOG_LEVEL=0 and rerun for more info.), with the frontend failing to connect to the server.

I had the same issue and after lots of fiddling I think I got it to work properly w/ the GPU.

I used Ubuntu 20.04 WSL and CUDA 11.3 (so it would match the version for pytorch exactly). CUDA install was standard as described on the official page (https://developer.nvidia.com/cuda-11.3.0-download-archive).

I downloaded the latest cuDNN version from the archive that matched the CUDA version (https://developer.nvidia.com/rdp/cudnn-archive). Specifically, I got the 'Library for Linux (x86_64)', and the 'Runtime Library', 'Development Library' and 'Samples' for 'Ubuntu 20.04 x86_64 (Deb)'. Not sure if they're all needed though. Installed the .deb packages normally and the tar file as described here (https://docs.nvidia.com/deeplearning/cudnn/install-guide/index.html).

I don't think you need to compile jaxlib from source if you follow the standard install instructions for 'pip installation: GPU (CUDA)'. I couldn't even get it to compile properly but the install worked fine.

From here I just followed the included instructions. After that the warning disappeared and everything seemed to be okay. Also make sure your drivers on Windows are up to date. Good luck!

@saharmor saharmor merged commit 7293607 into saharmor:main Nov 20, 2022
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

Successfully merging this pull request may close these issues.

5 participants