NVIDIA GPU Operator

Canonical Kubernetes Publisher

Platform:

Channel Revision Published Runs on
latest/beta 4 14 Feb 2024
Ubuntu 22.04 Ubuntu 20.04
latest/edge 1 13 Jan 2024
Ubuntu 22.04 Ubuntu 20.04
1.29/stable 4 14 Feb 2024
Ubuntu 22.04 Ubuntu 20.04
1.29/beta 4 14 Feb 2024
Ubuntu 22.04 Ubuntu 20.04
1.29/edge 1 13 Jan 2024
Ubuntu 22.04 Ubuntu 20.04 
juju deploy nvidia-gpu-operator --channel 1.29/stable

Learn to deploy on juju >

Learn about configurations >

  • image-registry | string

    Image registry for nvidia-gpu-operator charm images.

    The value set here will replace the host portion of each image URL in the release manifests. If unset, the default registry from upstream manifests will be used.

    example) juju config nvidia-gpu-operator image-registry=""

  • namespace | string

    Kubernetes namespace in which to install the GPU operator components including deployments, daemonsets, configmaps.

    If unset, "default" is assumed.

    NOTE: This can only be specified during initial charm deployment since some attributes of kubernetes resources are non-modifiable. The admin is responsible for ensuring the namespace exists prior to deployment.

  • nfd-worker-conf | string

    Default: sources: pci: # wokeignore:rule=whitelist deviceClassWhitelist: - "02" - "0200" - "0207" - "0300" - "0302" deviceLabelFields: - vendor

    Defines the Node Feature Discovery (NFD) worker configuration.

    Refer to the NVIDIA documentation for additional NFD configuration options: https://kubernetes-sigs.github.io/node-feature-discovery/v0.14/reference/worker-configuration-reference.html

    example) juju config nvidia-gpu-operator --reset nfd-worker-conf