pipeline
pipeline copied to clipboard
Install nvidia device driver on GPU nodes
Is your feature request related to a problem? Please describe. When I launch a cluster with GPU nodes, I still need to install the nvidia device driver manually.
Describe the solution you'd like to see Install the device driver automatically when the cluster has GPU nodes in it.
Describe alternatives you've considered Provide an option to install it (either during cluster create or as an integrated service)
Additional context GPU instances are not necessarily launched with GPU capable images. The device driver requires Docker and the nvidia drivers and docker runtime to be present. Without those, the device driver will fail to launch.