JupyterHub Multi GPU

Hi there,

I’ve got a cluster with 6 worker nodes, each with one gpu and I was wondering if it’s possible to use all 6 in one notebook? I know the pod for the notebook would be hosted on a particular node, so the architecture of Kubernetes might not allow it, but I wanted to ask, since I’m relatively new with Kubernetes and jupyterhub.


You’ll need another package that can distribute tasks across multiple nodes, for example

Any individual pod can’t use all the GPUs if they are located across multiple machines without using something like Dask that helps distribute work to different machines. But, if you have a single machine that has 6 GPU you can request 6 GPUs.