[Python] jupyter notebook with vscode on remote machine does not see GPU in tensorflow

I run vscode with a remote machine, using SSH. The tensorflow environment has tensorflow 2.17, python 3.11, and within vscode, python windows detect the GPUs . However, the same python code that detects the GPUs will not detect the GPUs when run in a Jupyter window in vscode.

This question appears to be the same as How to use a remote machine's GPU in jupyter notebook but the solutions there do not work in my case.

I found this: https://cschranz.medium.com/connect-vs-code-to-your-remote-gpu-jupyter-instance-58b86a195d9e

but I did not understand it. It looked like I need to run a separate jupyter instance, but again, I did not understand it and do not know if that would solve my problem.

Why does the jupyter windows produce a different result than the python window in vscode? What needs to be done to get Jupyter in vscode to see the GPUs?

My current workaround is to write the code in a python window, though this is less convenient than working with the Jupyter cells.

Continue reading...