AI workloads / Public reference

AI compute utilization

A supplied computeUtilization, gpuUtilization, or cpuUtilization of 85 percent or more.

What you need to supply

Telemetry you include, as a percent.

If it reports a finding

Medium below 95 percent, and high at or above that. The suggested next step is to review limits, batching, model size, and capacity.

What this does not prove

For an AI workload review, cpuUtilization is read as compute utilization. It is not the scaling replica check.

Telemetry field

{ "computeUtilization": 85 }

All review checks. Reading this page does not submit your infrastructure.