Same Cluster, 33 Points More Utilization: What Changed Was the Order

TL;DR - The title suggests a GPU workload-ordering change improved cluster utilization by 33 percentage points without adding hardware. No article content was provided, so implementation details cannot be verified.

• The reported gain comes from changing workload order, not cluster capacity.
• Scheduling policy can materially affect GPU utilization.
• Specific algorithms, benchmarks, and constraints are unavailable.