When monitoring a GPU-based workload, what is GPU utilization?

GPU utilization is defined as the percentage of time the GPU’s compute engines are actively processing data, reflecting its workload intensity over a period (e.g., via nvidia-smi). It’s distinct from memory usage (a separate metric), core counts, or maximum runtime, providing a direct measure of compute activity.

(Reference: NVIDIA AI Infrastructure and Operations Study Guide, Section on GPU Monitoring)