gpu monitoring

A 2026 field-by-field DCGM and AMD SMI telemetry crosswalk for GPU fleets, covering utilization, VRAM, energy, PCIe, health events, exporter versions, alert migration, and validation.

A 2026 operator guide to DCGM exporter GPU monitoring: field types, alert routing, MIG labels, PromQL templates, cardinality controls, and an incident triage checklist.

A 2026 data-driven guide to good GPU utilization percentages, covering NVIDIA's 80% SM Activity benchmark, Cast AI's 5% enterprise average, Model FLOPs Utilization, and how to monitor and raise GPU utilization.