GPU Instances
Manage GPU Instances
Control your GPU instance lifecycle — pause to save costs, resume when ready, or terminate when done.
Instance statuses
Every GPU instance is in one of the following states:
| Status | Description | Billing |
|---|---|---|
PENDING | Instance is being provisioned. Hardware is being allocated and your container image is loading. | No charge |
RUNNING | Instance is fully operational. SSH and JupyterLab are accessible. | Full compute rate (per-second) |
STOPPED | Instance is paused. GPU is released, but your Volume Disk and Network Volume data are preserved. | Reduced idle rate (storage only) |
TERMINATING | Instance is shutting down. Resources are being released. | Final settlement in progress |
TERMINATED | Instance has been fully shut down. All ephemeral storage is deleted. No further charges. | No charge |
FAILED | Provisioning or runtime error. The instance could not start or crashed unexpectedly. | No charge |
Pausing an instance
Click the Pause button on a running instance to stop it. The GPU is released immediately, and billing switches to a reduced idle rate that covers storage retention only.
- Your
/workspacedata (Volume Disk or Network Volume) is fully preserved. - Container Disk contents are also preserved — installed packages and OS-level changes survive the pause.
- SSH and JupyterLab become unavailable while paused.
Use pause when you want to keep your environment intact but aren't actively using compute. It's significantly cheaper than leaving the instance running.
Resuming an instance
Click Start on a stopped instance to resume it. The system allocates a GPU and restores your container with all previous data. This typically takes 15–60 seconds.
ℹ GPU migration: If the exact GPU model from your original deployment is unavailable in your region, you'll be offered the option to migrate to another available GPU. The instance rate adjusts to match the new hardware.
Terminating an instance
Click Terminate to permanently shut down an instance. This action:
- Stops all billing for this instance immediately (a final settlement is processed).
- Permanently deletes the Container Disk and Volume Disk.
- Does not delete any attached Network Volume — the volume remains available for future instances.
⚠ This is irreversible. If you haven't saved your work to a Network Volume or downloaded it, it will be lost permanently.
Billing lifecycle
Billing is calculated per-second and settled automatically on every status transition (deploy → running → paused → resumed → terminated). Your balance is debited for the exact seconds elapsed at each stage's rate. Check the Deploy page in your dashboard for current rates.