How do I investigate thermal throttling on a dedicated server?
Thermal throttling reduces performance to keep hardware within its supported operating limits. Low observed clock speed can also result from idle power policy, workload behaviour or a power cap, so temperature alone is not the whole diagnosis. Correlate sensors, workload and vendor-reported throttle reasons over the same interval.
Before you start
- Authorised access to the dedicated server and the named inspection tools.
Instructions
1. Read available operating-system sensors
Where lm-sensors is already configured, use its reporting command only. Sensor names and accuracy depend on the board and loaded drivers.
sensorsCompare readings with the hardware vendor’s thresholds. Do not run sensor-probing or limit-setting operations as part of a routine read-only check.
2. Compare management-controller health
If local IPMI is supported, inspect the platform’s temperature, fan and power-related readings.
sudo ipmitool -I open sensor listA missing sensor or an unavailable reading requires platform context. Combine the evidence with the affected CPU/GPU’s supported clock/throttling telemetry.
Check the result
- A sustained performance drop is tied to a specific thermal or power condition, and the provider confirms cooling, airflow and hardware health before any tuning change.
Troubleshooting
- Temperatures look normal after the incident
- Collect time-aligned monitoring during the workload; later idle readings may miss the peak.
- One sensor has a surprising value
- Confirm its unit, presence and vendor meaning before treating it as a failed component.
- More performance is requested
- Raising safety limits is not a substitute for correcting cooling or an unsuitable workload placement.
Official references
Use the documentation for your installed version when a flag, service name or platform feature differs.