Create a Cluster

Create an Instant Cluster on Together’s high-performance GPU clusters. With features like on-demand scaling, long-lived resizable high-bandwidth shared DC-local storage, Kubernetes and Slurm cluster flavors, a REST API, and Terraform support, you can run workloads flexibly without complex infrastructure management.

POST /compute/clusters

Python

Python

from together import Together

client = Together()

response = client.beta.clusters.create(
  cluster_name="my-gpu-cluster",
  region="us-central-8",
  gpu_type="H100_SXM",
  num_gpus=8,
  driver_version="CUDA_12_6_560",
  billint_type="ON_DEMAND",
)

print(response.cluster_id)

200

{
  "cluster_id": "<string>",
  "cluster_type": "KUBERNETES",
  "region": "<string>",
  "gpu_type": "H100_SXM",
  "cluster_name": "<string>",
  "duration_hours": 123,
  "driver_version": "CUDA_12_5_555",
  "volumes": [
    {
      "volume_id": "<string>",
      "volume_name": "<string>",
      "size_tib": 123,
      "status": "<string>"
    }
  ],
  "status": "WaitingForControlPlaneNodes",
  "control_plane_nodes": [
    {
      "node_id": "<string>",
      "node_name": "<string>",
      "status": "<string>",
      "host_name": "<string>",
      "num_cpu_cores": 123,
      "memory_gib": 123,
      "network": "<string>"
    }
  ],
  "gpu_worker_nodes": [
    {
      "node_id": "<string>",
      "node_name": "<string>",
      "status": "<string>",
      "host_name": "<string>",
      "num_cpu_cores": 123,
      "num_gpus": 123,
      "memory_gib": 123,
      "networks": [
        "<string>"
      ]
    }
  ],
  "kube_config": "<string>",
  "num_gpus": 123
}

Authorizations

Authorization · string · header · default: default · required

Bearer authentication header of the form Bearer <token>, where <token> is your auth token.

Body

application/json

GPU Cluster create request

region · enum<string> · required

Region to create the GPU cluster in. Valid values are us-central-8 and us-central-4.

Available options: us-central-8, us-central-4

gpu_type · enum<string> · required

Type of GPU to use in the cluster

Available options: H100_SXM, H200_SXM, RTX_6000_PCI, L40_PCIE, B200_SXM, H100_SXM_INF

num_gpus · integer · required

Number of GPUs to allocate in the cluster. This must be multiple of 8. For example, 8, 16 or 24

cluster_name · string · required

Name of the GPU cluster.

driver_version · enum<string> · required

NVIDIA driver version to use in the cluster.

Available options: CUDA_12_5_555, CUDA_12_6_560, CUDA_12_6_565, CUDA_12_8_570

billing_type · enum<string> · required

Available options: RESERVED, ON_DEMAND

cluster_type · enum<string>

Available options: KUBERNETES, SLURM

duration_days · integer

Duration in days to keep the cluster running.

shared_volume · object

Show child attributes

volume_id · string

Response

200 - application/json

OK

cluster_id · string · required

cluster_type · enum<string> · required

Available options: KUBERNETES, SLURM

region · string · required

gpu_type · enum<string> · required

Available options: H100_SXM, H200_SXM, RTX_6000_PCI, L40_PCIE, B200_SXM, H100_SXM_INF

cluster_name · string · required

duration_hours · integer · required

driver_version · enum<string> · required

Available options: CUDA_12_5_555, CUDA_12_6_560, CUDA_12_6_565, CUDA_12_8_570

volumes · object[] · required

Show child attributes

status · enum<string> · required

Current status of the GPU cluster.

Available options: WaitingForControlPlaneNodes, WaitingForDataPlaneNodes, WaitingForSubnet, WaitingForSharedVolume, InstallingDrivers, RunningAcceptanceTests, Paused, OnDemandComputePaused, Ready, Degraded, Deleting

control_plane_nodes · object[] · required

Show child attributes

gpu_worker_nodes · object[] · required

Show child attributes

kube_config · string · required

num_gpus · integer · required