Skip to main content

Launch Compute

Use compute when you need direct GPU or CPU capacity instead of a packaged app experience. Compute runtimes are project-scoped, billable, and should be released when work is complete.

Flow​

Launch From The UI​

  1. Confirm the tenant and project in the top bar.
  2. Open Compute.
  3. Choose the SKU or slice profile.
  4. Choose the image or runtime template.
  5. Select the SSH key or supported access method.
  6. Review hourly burn and quota.
  7. Launch.

Compute launch form

Launch Inputs​

InputWhy it matters
ProjectOwns membership, billing, and runtime visibility.
SKU or sliceDetermines GPU/CPU/memory capability and hourly burn.
RegionDetermines capacity pool and route locality.
ImageDefines the runtime base environment.
SSH keyEnables SSH-based access when supported.
StorageProvides workspace, model, or dataset access.

Connect​

Open the workload detail page after the runtime is active.

Supported compute connect actions can include:

  • SSH command;
  • browser terminal;
  • metrics;
  • workload detail and lifecycle actions.

If a connect action is missing, check:

  1. the runtime is active;
  2. your project role has access;
  3. your SSH key is registered if SSH is required;
  4. the route or terminal session has not expired;
  5. no warning or needs-action item is visible.

Monitor Usage​

During a run, review:

  • runtime state;
  • hourly burn;
  • project balance or quota;
  • active session state;
  • metrics or health if available.

Long-running compute should have an owner and a reason. Avoid leaving capacity running after validation.

Release​

Release from the workload detail page when finished. Release should stop billing for the runtime and move cleanup forward.

If release fails:

  • use the visible retry when available;
  • capture the correlation ID;
  • do not manually reuse the runtime for new work until cleanup state is clear.

CLI Equivalent​

Use the CLI for state checks and cleanup automation.

gpuaas allocations list --status active --output table
gpuaas allocations release --id <allocation_id> --project-id <project_id>
gpuaas allocations list --status released --project-id <project_id>

For browser terminal or SSH launch details, use the connect action shown on the workload detail page.

Support Packet​

When asking for help, include:

  • tenant and project name;
  • runtime name or allocation ID;
  • visible state;
  • action attempted;
  • correlation ID;
  • whether SSH, terminal, route, or release failed.

Do not include private keys, bearer tokens, one-time codes, cookies, or raw session tokens.