Why should GPU utilization be interpreted alongside memory, throughput and application metrics?

1. High utilization always proves useful work
2. Memory usage alone guarantees maximum throughput
3. Low utilization always means failed hardware
4. A single utilization percentage cannot explain end-to-end efficiency