Inventory
Captured at enrollment:- Driver and CUDA versions
- Every GPU’s UUID — the identity anchor for hardware-change detection
- VRAM and MIG mode
- The interconnect topology matrix
Per-GPU live charts
Utilization, VRAM, temperature, and power — plus ECC single/double-bit and Xid lifetime counters.ECC and Xid counters are the numbers that tell you whether the silicon is
healthy. They are lifetime counters, so what matters is the rate of change
against a known-good baseline, not the absolute value on a card that has been
running for a year.
Conditions
Agent-reported health checks with reasons:- NVML availability
- Clock skew
- Buffer pressure
Log viewer
Search by time range, level, source, or free text. Follow mode tails live. Xid and NVRM lines are highlighted.Hardware changes
Because identity is the UUID rather than a slot index, a card moved between
slots is not reported as a swap, and a card silently replaced with a different
one cannot pass as the original.