rp cluster create

Provision a homogeneous multi-node cluster.

rp cluster create --name <n> --type <kind> --gpu <type>
                        [--pod-count N] [--gpu-count N] [flags]

Options

  --name <n>                        cluster name (required)
  --type <kind>                     APPLICATION|TRAINING|SLURM|RAY (required)
  --gpu <type>                      one GPU type for every pod (required)
  --gpu-count N                     GPUs per pod (default: 1; minimum 1)
  --pod-count N                     number of pods (minimum 2, default: 2)
  --dc <id,…>                       preferred datacentres; omit to let the
                                   scheduler place the cluster (single DC)
  --image <ref>                     Docker image for every pod
  --container-disk-gb N             ephemeral container disk, GB (minimum 1)
  --ports <a/b,…>                   exposed ports, each as port/protocol
  --env K=V                         environment variable; repeatable
  --start-cmd <a,b,…>               arguments passed to the container entrypoint
  --network-volume-id <id>          attach one network volume to every pod
  --volume-path <path>              mount path for the network volume
  --template-id <id>                seed container config from a template id
  --start-ssh true|false            provision SSH with your account key
  --start-jupyter true|false       start Jupyter on every member pod
  --force                           create even when the name is taken

Notes

A cluster is homogeneous: every pod is identical, so a single --gpu type, --gpu-count, and --pod-count describe the whole fleet. --pod-count must be at least 2 (a cluster is multi-node by definition). Creation is idempotent by name, like rp volume create and rp serverless create: where a cluster of that name already exists, the CLI prints its id and skips the POST. --force sends the request regardless. Only the name can change afterwards — rp cluster update is a rename, and compute shape, type, and container config are fixed at create. Clusters do not yet support private registries (no --registry): the v2 create request has no registry field. Use a public image or a network volume for your build. --template-id seeds the container config as defaults; any explicit flag value still wins, and the template's id is recorded on the cluster.

Examples

# Create a 4-node H100 training cluster
$ rp cluster create --name tr-1 --type TRAINING \
      --gpu "NVIDIA H100 80GB HBM3" --pod-count 4 --gpu-count 8

# Create a Ray cluster with an attached volume
$ rp cluster create --name ray --type RAY --gpu "NVIDIA L4" \
      --image runpod/ray:latest --network-volume-id vol_xyz

API: POST /v2/clusters