NVIDIA logo

Skill

dgx-station-mig

manage NVIDIA DGX Station MIG profiles

Covers Operations NVIDIA Engineering

Description

Inspect NVIDIA DGX Station GB300 MIG state and installed-driver profiles, create a digest-bound layout plan, disclose disruption and restoration, and apply an approved still-valid plan. Use when the user asks to enable, disable, partition, reconfigure, inspect, or troubleshoot MIG instances or needs MIG UUIDs. Never assume static profile IDs or terminate GPU clients.

SKILL.md

DGX Station MIG

Treat inspection and planning as read-only. Treat every apply as disruptive.

Workflow

  1. Run scripts/dgx-assist system inspect --json.
  2. Search the playbooks for the user's MIG concern and cite relevant results.
  3. Read compatibility.capabilities.mig_inspection. If it is false, stop and explain that the detected release has no MIG inspection profile.
  4. Run mig inspect and mig profiles. Use only profiles reported by the installed driver. This read-only step is available on the recognized Software 1.0 and Software 2.0 profiles.
  5. Before planning, require compatibility.capabilities.mig_mutation to be true. If it is false, report the observed state but do not suggest a layout or work around the restriction.
  6. Require the desired exact layout. Do not choose a layout from model names or nominal memory sums.
  7. Run mig plan --layout "<layout>".
  8. Present current and proposed state, exact commands, active clients, privilege, workload impact, reset/reboot possibility, and restoration commands.
  9. If any client is active, stop. Tell the user dgx-assist will not terminate it.
  10. Obtain explicit confirmation immediately before mutation.
  11. Run mig apply --plan-id "<id>" --yes.
  12. Report every verified MIG UUID. On partial failure, stop and present only the recorded restoration plan.

Safety requirements

  • Never use a hardcoded GPU index, profile table, or assumed placement.
  • Never destroy instances, reset a GPU, reboot, or change MIG mode outside an approved plan.
  • Never kill a GPU process or take over a service.
  • Invalidate the plan when clients, mode, profiles, release, driver, or instances change.
  • Do not improvise further mutations after partial failure.
  • Do not treat Fabric Manager as a normal Station prerequisite.
  • Treat --yes only as approval already obtained.

Read references/workflow.md before planning or applying a layout.

© 2026 YourAI.tools. Every skill from an identity-verified publisher.

Independent catalog. Not affiliated with, endorsed by, or sponsored by Anthropic or any listed publisher. All trademarks belong to their respective owners.