GPU utilization in a Terra notebook

Post author
Sabrina Camp

Hi! 

I was working through [this](https://support.terra.bio/hc/en-us/articles/4403006001947-Getting-started-with-GPUs-in-a-Cloud-Environment) documentation on using GPUs in a Terra notebook. 

This is the set up of my application configuration.

 

I was trying to verify that I've enabled GPUs by using the pytorch commands given in the doc. Since I didn't select an application configuration with pytorch already installed, I first installed pytorch via `pip install torch`. 

Once successfully installed and imported, I attempted the example commands and it doesn't seem like it's working properly. 

 

Any advice? Is it necessarily to use the pegasus environment to work with the GPUs?

 

Thanks!

Sabrina Camp

 

Comments

2 comments

  • Comment author
    Samantha (she/her)

    Hi Sabrina Camp,

    Thanks for writing in. Unfortunately, it looks like using GPUs in Jupyter Notebooks is currently broken. Our engineers are aware and have a ticket to resolve the issue.

    In the meantime, one workaround that was identified is to downgrade CUDA by running the following:

    # install CUDA 11.8
    !pip install torch==2.1.2 torchvision==0.16.2 torchaudio==2.1.2 --index-url https://download.pytorch.org/whl/cu118

    Once you restart the kernel, you should be able to run those example commands again.

     

    Best,

    Samantha

    0
  • Comment author
    Sabrina Camp

    Hi Samantha,

    This workaround worked for me. 

    Thank you,

    Sabrina

    0

Please sign in to leave a comment.