EdgeNVIDIA

MicroK8s

Canonical

A small, single-package Kubernetes ideal for dev and edge. A GPU addon enables the NVIDIA GPU Operator with one command.

Bootstrap with MicroK8s
Enablement

microk8s enable gpu (GPU Operator addon)

Driver install

The addon deploys the GPU Operator, which installs drivers.

Container runtime

containerd (bundled).

Scheduling

Default scheduler; typically single-node or small clusters.

Partitioning

Time-slicing for sharing a card across dev workloads.

Networking

Calico (default); minimal for edge.

Upgrades

Snap channel refresh; re-enable/verify the GPU addon after upgrades.

GPU enablement
  1. 1

    Enable the GPU addon

    Deploys NFD + the GPU Operator automatically.

    bash
    microk8s enable gpu
  2. 2

    Wait for the operator

    Give the operator time to install drivers and the device plugin.

    bash
    microk8s kubectl -n gpu-operator-resources get pods
  3. 3

    Run a GPU workload

    Request nvidia.com/gpu in a test pod.

Validate the enablement
verify GPUs
microk8s kubectl get nodes -o json | jq '.items[].status.allocatable' | grep gpu
microk8s kubectl -n gpu-operator-resources get pods
Known limitations
  • ·Aimed at dev/edge, not large multi-node training.
  • ·Single-node by default; scale-out needs clustering.
Best for

Developer sandboxes and edge sites needing a tiny, quick GPU cluster.