> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://docs.nvidia.com/ccluster/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://docs.nvidia.com/ccluster/_mcp/server.

# Compute Instance

> Launch an on-demand GPU instance from any container image and start building

On-demand GPU instances with SSH access to run any GPU-accelerated workloads and let you run, test, and experiment with any AI application seamlessly.

## 1. Choose an image

Compute instances can run any container image. In the **Add or Select Image** field, type the name of an image from a container registry and set the **Container Tag**. For example, use image `ubuntu` with tag `24.04`, or image `nvcr.io/nvidia/pytorch` with tag `26.01-py3`. If your environment defines suggested images, they appear in the dropdown.

On full GPU instances, NVIDIA drivers are included via driver passthrough, so the GPU is accessible even from a minimal image like `ubuntu`. On MIG instances, NVIDIA drivers are **not** available yet. CUDA and framework libraries are not injected and must come from the image itself, so for GPU workloads pick an image that bundles them, like the [NVIDIA NGC](https://catalog.ngc.nvidia.com/containers) PyTorch image.

> **Note**
>
> Your image does not need an SSH server preinstalled. CCluster injects one at deploy time without modifying the image. The SSH server runs as the container's main process, so the image's own `ENTRYPOINT`/`CMD` is not executed. Connect over SSH to start your workloads.

### Private registry images

To deploy an image from a private registry (e.g., Amazon ECR, Google Artifact Registry, Azure Container Registry, or a self-hosted registry), provide credentials in the **Image Registry Username** and **Image Registry Password** fields. CCluster automatically detects the registry from the image URL. For public images, leave both fields empty.

API deployments can instead [reference an existing image pull secret](/ccluster/resources/custom-image#use-an-existing-image-pull-secret) managed by your cluster administrator using `image_pull_secret_name`.

## 2. Configure access and deploy

Enter your SSH public key to configure access to the instance, select a GPU instance type, and click Deploy.

Optional settings on the create form:

* **Enable logging** — collect container output and show it on the deployment **Logs** tab.
* **Annotations** — optional key/value metadata on the instance.

![Compute instance launch page](/ccluster/_fern-files/_img/78383ae1aeb1ce77602e01e2e7b5722f40fb8462ec90be74b88d3b5ac1b7eef8.webp)

## 3. SSH into the instance

Once the instance is ready, navigate to the deployment details page. The **Endpoint Configuration** section displays:

* **Endpoint URL** — the hostname for your instance. Next to it are the copy button (copies the URL) and the SSH button (copies `ssh root@<endpoint_url>` so you can paste it directly into your terminal).

![Endpoint Configuration section showing the endpoint URL and SSH public key](/ccluster/_fern-files/_img/3c5fb27d5c4401bece6971117e8abe74a65acf01e1d068adfe1ec2bebb08829f.webp)

To connect, use the SSH command with the `root` user:

```bash showLineNumbers={false}
ssh root@<endpoint_url>
```

![Running compute instance showing its endpoint URL](/ccluster/_fern-files/_img/968142b37e11d8e67e47b61e3ab3b2bb44379e6bac0c4f5abd2e88108a2f5b89.webp)

The instance authenticates you with the SSH public key you provided at deployment. You can also run one-off commands, transfer files, and forward TCP ports:

```bash showLineNumbers={false}
# Run a one-off command
ssh root@<endpoint_url> 'nvidia-smi'

# Copy files to and from the instance with scp or SFTP
scp ./train.py root@<endpoint_url>:/home/
scp root@<endpoint_url>:/home/results.csv .

# Forward a port from the instance to your local machine (TensorBoard in this example)
ssh -L 6006:localhost:6006 root@<endpoint_url>
```

X11 forwarding and SSH agent forwarding are not supported.

The interactive shell depends on what your image provides:

| Image                                                 | Shell over SSH                                       |
| ----------------------------------------------------- | ---------------------------------------------------- |
| Images with `bash` (e.g., Ubuntu, Debian, NGC images) | `bash` login shell                                   |
| Images with only `sh` (e.g., Alpine)                  | `/bin/sh`                                            |
| Images without a shell (e.g., distroless)             | No interactive shell. SFTP file transfer still works |

The instance includes the libraries from your image. Additional packages can be installed with your preferred package manager.

### In-console Terminal

While the instance is **Active**, the deployment **Terminal** tab opens a shell in the browser without SSH from your machine. Use SSH (above) for `scp`/`SFTP` and port forwarding. See [Managing Deployments](/ccluster/resources/managing-deployments) for Terminal availability, pause/resume, clone, and logs.

Compute deployments are not revision-aware in the console: edits update the instance in place and there is no rollback table.

## What's next

#### [Managing Deployments](/ccluster/resources/managing-deployments)

In-console Terminal, logs, pause, and clone.

#### [Managing Vault Objects](/ccluster/resources/vault)

Store SSH keys and registry credentials for reuse across deployments

#### [LLM Serving](/ccluster/apps/llm)

Explore dedicated, protected endpoints for production model deployments.