Monthly allocation
Share of total possible capacity in each month; allocation-hours are shown as supporting values.
CPU capacity allocated
0–100% · months on x-axisHow long jobs wait in the queue before starting
Daily average wait, separated by requested GPU count.
Job runtime distribution
Started jobs grouped by elapsed runtime.
Queue-wait distribution
Started jobs grouped by submit-to-start delay.
Largest GPU Units
Period total and share of all GPU-hours.
Largest CPU Units
Period total and share of all CPU-hours.
Notable GPU workloads
Largest assigned projects by GPU-hours.
Notable CPU workloads
Largest assigned projects by CPU-hours.
How the Unit mix changed
Monthly percentage of the corresponding cluster total. Top six Units are shown separately.
Detailed Unit allocation
Every value includes hours and percentage of that column’s total.
Historical snapshot
The preserved September–October 2025 report provides an earlier point of comparison.
Method
- Allocation, not utilization: requested GPU or CPU count × elapsed running time within each month.
- Running jobs are counted only through the stated cutoff. Pending time is excluded.
- All unique monitoring records are retained, including valid requeues and reused job IDs.
- Unit mapping uses time-bounded monitoring assignments. Any unmapped interval is reported as Unassigned.
Interpretation notes
Unit percentages use all cluster allocation as the denominator, including Unassigned activity. CPU-hours and GPU-hours describe different resources and should not be combined into a single utilization score.
The two headline capacity percentages assume 62 H100 GPUs and 1,792 logical CPUs were continuously available throughout the reporting period. They measure allocation intensity, not device utilization or efficiency.