Skip to content

Stop restart and destroy a GPU rental safely

Protect workspace data and understand stopped-storage costs before ending a rental.

Before you start

Save running job output and download important files. Stopping, restarting, and destroying have different effects.

Steps

  1. Open the GPU instance and inspect its status and costs.
  2. Stop the instance when you want to pause compute while retaining supported rental storage. Confirm that the provider supports stop/start for the rental.
  3. Start the existing rental to resume it. Check that models and workspace files remain and that the application is ready.
  4. Use restart only after saving work; it interrupts the application even when storage remains.
  5. To end the rental permanently, back up required files and choose Destroy or the equivalent termination action. Confirm the displayed deletion warning.
  6. Verify the terminated state and review wallet usage. Destroying a rental removes its storage; keep long-term backups outside it.

Check the result

The intended state is reflected in the instance page. A stopped rental can still incur disk charges; a destroyed rental is not a recovery point.

Troubleshooting

Provider capacity can affect restarting a stopped rental. Never assume a stopped disk is a backup or that terminating compute cancels unrelated subscriptions.

Last updated October 12, 2026. Availability and prices are shown in the console before checkout.
Still need help? Open a support ticket with your service name, the time of the issue, and a redacted error message. Never include passwords or API keys.