Infrastructure
Track thermal state, utilization, power draw, idle opportunity, and current allocation in one operational view.
Each block updates independently while preserving its workload state.
Temperature trend
Cluster average temperature over recent cycles
Utilization trend
Average useful GPU activity across the fleet
Operational thresholds
The existing backend classification remains the source of truth.
Low
Below 60°C
Normal thermal state
Medium
60–74°C
Monitor placement
High
75–84°C
Move workload
Critical
85°C and above
Safety override