Is there a way to connect a JupyterHub spawned notebook that runs within a kubernetes cluster to multiple GPUs that are located:

a) on the same worker node

b) on other nodes

Thank you for your ideas.