mediumHardware & health
Thermal throttling
Clocks drop under load because the GPU is too hot.
Scope
Cooling/facility problem manifesting as lost throughput.
Check recent changes
- ·Higher ambient temp
- ·Denser packing
- ·Cooling maintenance
Confirm the throttle
bash
nvidia-smi -q -d PERFORMANCE,TEMPERATURE | grep -Ei 'throttle|temp'Expect: 'SW Thermal Slowdown: Active' with high temperature.
- high likelihoodInadequate cooling for the power density
Fix · Improve airflow / move to liquid cooling for 700W+ parts.
- medium likelihoodFailed fan or pump
Fix · Replace the cooling component.
- low likelihoodHigh ambient temperature
Fix · Engage facility cooling.
Validation
- ✓Temperature under the throttle threshold; clocks hold at load
Escalation
Facility-wide cooling shortfall → data-center operations.
Data to collect first
- ·Temperature vs. clock
- ·Per-rack thermals
- ·Fan/pump status
Related alerts