Limit GPU cores for training

My company has a 6-gpu server. However, I only want to train the model on a specific number of gpus (maybe 1 or 2).
How can I achieve that?

What are you using for training ? Trainer or the pytorch-lightning ?

You can set this env variable

CUDA_VISIBLE_DEVICES=0,1

specify the devices that you want to use and the Trainer will only use those cuda devices.

For pytorch-lightninig here’s official doc

1 Like

Thank you. This is exactly what I need

Adding these two lines at the top of a Notebook (or Python script) can limit the devices to use for that script.

import os
os.environ["CUDA_VISIBLE_DEVICES"]="0,2"