Cluster dashboard

Press d from the main screen for a cluster-wide overview, refreshed every 60 s from two sinfo calls.

Cluster dashboard

How to interpret each row

  • Capacity — sum across all nodes. CPU and memory percentages are computed from the per-node %C (Allocated/Idle/Other/Total) and RealMemory - FreeMem fields.

  • Partition table columns:

    • Nodes (i/m/a/d) — counts by state (idle / mixed / allocated / down|drain|fail).

    • CPUs (free/total) — total - alloc per partition.

    • GPUs (used/total) — counted from the partition’s Gres field plus per-node usage.

    • Mem — the largest RealMemory value seen for the partition, so heterogeneous partitions still show a meaningful number rather than double-counting.

    • Up — derived from sinfo’s %a Availability flag.

  • Node table — the per-node rollup, color-coded by state.

Note

GPU usage is approximate: allocated / mixed nodes count as “all GPUs used” because sinfo does not expose per-job GPU counts. For exact per-job GPU utilisation, open the Job detail screen — it queries nvidia-smi inside the job’s allocation.

Keybindings

Key

Action

r

Force-refresh sinfo now.

j / k / g / G

Scroll the body.

?

Help cheatsheet.

Esc / q

Back to the job list.