Capacity & Storage

What the cluster is made of, and how much data it holds.

Data freshness unknown

Peak performance not yetAggregate dense FP16 tensor throughput. No sparsity multiplier applied.

GPUs installed not yet

GPUs online not yetCurrently up and schedulable.

Storage not yetProvisioned across all shared filesystems.

Hardware

GPU mix

Discovered live from the scheduler, not read from a config file — so this cannot drift from reality.

Storage

Capacity by filesystem

Used and free space across project and scratch filesystems. Scratch is purged on a rolling window; project storage is persistent.

Storage I/O (yesterday)

Total bytes read and written across the shared storage nodes in the most recent complete calendar day, integrated from Prometheus rate series. See the methodology page for why cumulative counters from zpool iostat are not used.

Peak PFLOPS is a theoretical ceiling computed from vendor specifications for the installed hardware, using dense (non-sparse) tensor-core figures. It is a description of the machine, not a benchmark result, and no real workload reaches it.