Get started with GPU Node

Launch a GPU node, sign in to it, and see its card listed, ready for your first job. Before you start, have a MyAccount account with credit or a payment method on file, and set your region to Delhi, where GPU nodes are offered.

Create a GPU node

Open Compute > Nodes, click Add New Node, and switch from the CPU tab to the GPU tab. Pick a card, and you land on Select Plan.

For a first node, pick a card with enough GPU memory for what you want to try; each plan shows its GPU memory, and the Summary panel shows the price before you launch. See Choose a GPU card once you are picking a card for a real workload, the job, service, or model you actually plan to run.

On Select Plan, choose the operating system, version, and plan the same way as a regular node; each plan shows vCPU count, RAM, GPU memory, and CUDA version so you can compare before you pick. Leave the optional settings below the plan at their defaults for your first node, add an SSH key or leave it to get an emailed password, then review the Summary panel and click Launch.

Launching takes you to your node list, where the new node shows as Creating and then Running.

Connect and check the GPU card

Once the node reads Running, connect to it the same way as any node:

ssh <username>@<your-node-ip>

Find the node’s IP on its details page. If you used a password instead of an SSH key, the username and password are in the email E2E sent you when the node was created. Once you are in, run:

nvidia-smi

A healthy node prints the card name, driver version, CUDA version, memory, and utilization. If the command is not found or reports it cannot reach the driver, see Troubleshooting GPU Node.

What to do next with your GPU node

• Choose a GPU card: what each card is built for, so your next node starts from the right one.
• Run GPU workloads in Docker: the recommended way to run a real job once the card checks out.
• Create a GPU node: every field this quickstart skipped, including billing and network options.