2 years ago

Have 1 or several VM/servers with at least 1 GPU if you want to deploy a multi-node Kubernetes cluster, the master node root user should be able to ssh on each worker. Create a k0s Kubernetes cluster https://github.com/k0sproject/k0s is a simple tool to deploy a Kubernetes cluster.

Then you need to create a k0sctl.yaml config file: For a single node Kubernetes cluster apiVersion: k0sctl.k0sproject.io/v1beta1 kind: Cluster metadata: name: my-cluster spec: hosts: - role: single uploadBinary: true k0sBinaryPath: /tmp/k0s localhost: enabled: true files: - name: containerd-config src: /tmp/containerd.toml dstDir: /etc/k0s/ perm: "0755" dirPerm: null k0s: version: 1.24.4+k0s.0 config: spec: network: provider: calico

Deploy NVIDIA GPU Operator https://github.com/NVIDIA/gpu-operator is a Kubernetes tool that manages NVIDIA plugin installation on your cluster nodes.

# Add the label to all the cluster nodes k0s kubectl get nodes -o name | xargs -I {} k0s kubectl label {} gpushare=true