During a municipal facility power constraint, how can a UCS power policy help GPU servers?

During a municipal facility power constraint, a UCS power policy helps GPU servers stay operational by capping or prioritizing GPU server power consumption so collective draws stay within facility limits. The policy ensures trainers get their needed power while non-critical infrastructure backs down—a clean priority order that keeps the GPUs running without a facility-wide blackout. Attaching a power policy that enables N+1 PSU redundancy for GPU servers only improves availability against a PSU failure; it does not cap or prioritize GPU server power draw, so it cannot keep collective consumption within facility limits and risks a cascading blackout. Attaching a power policy that schedules GPU server power capping for off-peak maintenance windows only restricts power during a future maintenance period; it does not cap or prioritize power during the active constraint, so GPUs remain free to exceed facility capacity. Attaching a power policy that caps CPU socket power domains while leaving GPU power domains unrestricted leaves the dominant GPU power draw uncontrolled; it does not cap or prioritize total server power, so it is not an effective power management solution for an active facility constraint. Exam caveat: power policy must be tested under worst-case load (all GPUs max power) to verify the facility constraints are actually respected; a misconfigured policy may allow oversubscription. Operational check: configure a UCS power policy that prioritizes GPU servers and caps other loads, simulate peak collective power draw, and verify facility power consumption stays below the constraint and GPUs receive sufficient power to run at full performance.