Clusters Quick Start

Get your first GPU cluster running in minutes

Clusters are ideal for interactive development, debugging, and building new projects. If you have working code and want to run experiments, Jobs are recommended instead.

Make sure you’ve installed the TensorPool CLI and configured your API key.

1. Create Your First GPU Cluster

   Create a 1xB200

   tp cluster create 1xB200
   

   For multi-node training, create a 4-node 8xB200 cluster:

   tp cluster create 8xB200 -n 4
   

   You can also create a cluster with a pre-built container image (includes CUDA, Python, and ML libraries):

   tp cluster create 1xB200 --container pytorch
   

   See instance types for all available GPU configurations and container images for available images.

2. Check Your Cluster Status

   The tp cluster create command will give you a cluster ID (e.g., c-abc123). Use it to check your cluster’s status:

   tp cluster info <cluster_id>
   

   Wait until the status shows RUNNING. The output will list your cluster’s instances, each with an instance ID (e.g., i-xyz789)

   If you lose the cluster ID, you can always find it with tp cluster list

3. SSH Into Your Cluster

   Once your cluster status is RUNNING, grab the instance ID from tp cluster info and connect:

   tp ssh <instance_id>
   

   For multi-node clusters, SSH into the jumphost instance first. From there, you can access worker nodes by name (e.g., ssh <cluster_id>-0).

4. Clean Up

   When you’re done, destroy your cluster:

   tp cluster destroy <cluster_id>
   

Next Steps

• Learn about cluster management
• Explore storage volumes for persistent data
• Check out multi-node training for distributed workloads