GPU Notebook for CNNs

Is there a way to create a Python environment with a GPU aviable? I need to run a neural network on the GPU using the tensorflow package.

1 Like

Hi @Manuel_Trevisan - that’s a request we hear about frequently. We don’t have a formal plan yet, but we are looking at ways to allow you to change machine configurations more easily.

Have you tried using Colab for this?


Wait, so you’ve figured out how to get tensorflow running? Can you help me with that?

Yes, but on Colab you have a limited amount of time if you don’t buy a plan. Also, in Project IDX, you can set up your environment in a better way. It would be great if the pioneers of Project IDX could experiment using those resources as well.

Actually, it’s quite easy. Once you start your Gemini Python notebook environment (which should already have pip installed), just type pip install tensorflow in the terminal or !pip install tensorflow in a cell, and everything should work. However, I recommend running things on other platforms or on your own machine using the GPU, as it is pretty slow in IDX.

I’ve tried pip. It says that it isn’t allowed to modify system variables. I tried using pkgs.python311packages.tensorflow in the packages list but that doesn’t work, even if it successfully builds.

+1 for this idea
Although tools like colab and Kaggle exist, both of them do not have as much customizability as Project IDX
(and also, Project IDX’s autocomplete, debugger and other tools will always be superiors as vscode is arguably an IDE)