Deploy a custom Docker image on a GPU rental

Before you start

Your image must support the selected GPU, host NVIDIA driver, CPU architecture, and provider runtime. Ensure the provider can pull it; a private image needs a supported authentication path.

Steps

1. Select an offering and choose Custom Image / Manual Config on Configure.
2. Enter a pullable Docker image reference, preferably an immutable digest. Use the startup command or entrypoint required by that image.
3. Set disk sizes and a persistent mount path. Keep important data in the rental’s persistent workspace rather than an undocumented temporary path.
4. Add each application’s container port and choose the correct TCP or UDP transport and HTTP, HTTPS backend, or raw connection type where offered.
5. Configure SSH only if your image runs an SSH server with the expected user, key handling, and port. Review and deploy.
6. Inspect Container logs and Connections to verify the actual process and listeners. Add authentication inside publicly exposed applications.

Check the result

Your application runs using its own configuration and the published endpoints reach the correct container listeners.

Troubleshooting

A running container is not proof that an app is listening. Check image pull failures, startup exits, CUDA compatibility, disk space, and whether the service binds to 0.0.0.0 rather than localhost.