Use JupyterLab on a GPU rental

Before you start

Select a JupyterLab template compatible with your GPU and provider. Wait for startup to finish.

Steps

1. Open the GPU instance’s Connections section and launch the JupyterLab endpoint.
2. Use the application credential shown for the rental when prompted. Do not substitute your console API key.
3. Create or upload a notebook under the persistent workspace directory.
4. Choose the intended Python kernel and run a small torch.cuda.is_available() check if PyTorch is included.
5. Save the notebook and export important outputs before stopping or destroying the rental. Shut down unused kernels to free GPU memory.

Check the result

The notebook saves successfully and the kernel can access the assigned GPU.

Troubleshooting

If the browser loads but the kernel fails, inspect logs, kernel packages, memory pressure, and disk space. A notebook stored outside the configured persistent mount may be lost when the container is replaced.