Quickstart

​

Create a cluster

​

1. Access the cluster console

1. Log into api.together.ai
2. Select GPU Clusters in the top navigation menu
3. Select Create Cluster

​

2. Choose capacity type

• Reserved: Pay upfront to reserve capacity for 1-90 days with discounted pricing.
• On-demand: Pay hourly with no commitment. Terminate anytime.

​

3. Configure your cluster

• Select the number and type of GPUs (for example, 8xH100).
• Available options: H100, H200, and B200.

• Enter a descriptive name.

• Kubernetes: For containerized workloads and Kubernetes-native tools.
• Slurm: For HPC-style batch scheduling and traditional workflows.

• Defaults to Any region. Together assigns the region with the most available capacity for your selected GPU type when you create the cluster.
• Select a specific datacenter region instead if you need the cluster in a particular location.
• Changing the GPU type resets the region to Any region and clears any selected shared volume, because volumes are region-specific.

• Choose reservation length: 1-90 days.

• Create and name your persistent storage volume.
• Minimum size: 1 TiB.
• Can be resized later as needed.

• Select NVIDIA driver version.
• Select CUDA version.

​

4. Create and verify

1. Select Proceed to create your cluster.
2. Monitor the cluster status in the UI as it provisions.
3. Wait for status to transition to Ready.

​

Connect to the cluster

​

For Kubernetes clusters

1. Install kubectl
   • macOS installation guide.
   • Or use your preferred method for your OS.
2. Download kubeconfig Use the Together CLI to download the cluster’s credentials to your local ~/.kube/config. Find your cluster ID with tg beta clusters list:

tg beta clusters get-credentials <CLUSTER_ID> --set-default-context

3. Verify connectivity

kubectl get nodes

4. Start using your cluster
   • Deploy workloads.
   • Access the Kubernetes dashboard.

​

For Slurm clusters

1. Choose an SSH access method
   • On clusters with OIDC enabled, select OIDC or Key-based in SSH access method on the cluster details page.
   • OIDC: Install the Together CLI (uv tool install "together[cli]", requires CLI 2.20+ and Python 3.10+) and choose a login name when prompted. No SSH key is required.
   • Key-based: Add your SSH key at api.together.ai/settings/ssh-key before cluster creation.
2. Connect via SSH
   • Copy the head node command from the cluster sidebar with Copy head node SSH command.
   • Paste and run the command in your terminal to reach the Slurm login node.
3. Verify Slurm

sinfo          # View node status
squeue         # View job queue

4. Start submitting jobs
   • Learn about Slurm commands.
   • Submit batch jobs with sbatch.
   • Run interactive jobs with srun.
5. Optional: Download kubeconfig Slurm clusters run on Kubernetes, so if you need kubectl access to the underlying cluster API, any project member can download the kubeconfig from the cluster details page or with the Together CLI:

tg beta clusters get-credentials <CLUSTER_ID> --set-default-context

​

Common first tasks

​

Upload data

# Create a pod with your shared volume mounted
# Then copy files directly
kubectl cp local_file.tar.gz pod-name:/mnt/shared/

​

Run a test job

kubectl run test --image=ubuntu --command -- sleep infinity
kubectl exec -it test -- bash

srun --gpus=1 --pty bash
nvidia-smi

​

Troubleshooting

​

Can’t see my nodes

• Check cluster status in the UI (should be Ready).
• Re-download the latest credentials with tg beta clusters get-credentials <CLUSTER_ID>.

​

SSH connection refused

• Verify your SSH key was added before cluster creation.
• Check the connection command in the cluster UI.
• Ensure you’re using the correct hostname.

​

Capacity unavailable

• Use the Notify Me option to get alerts when capacity is available.
• Try a different region.
• Contact [email protected] for custom requirements.

​

Next steps