Purpose

This article provides guidance for engineers to identify and respond to the GPU Temperature Critical alert in GPU compute nodes. It outlines the alert meaning, possible causes, and the required L1 and L2 troubleshooting steps.

Alert Name

GPU Temperature Critical

Alert Description

This alert is triggered when the GPU temperature exceeds 85°C.

High GPU temperature typically occurs due to heavy workloads, insufficient cooling, or hardware issues affecting airflow.

Severity

P3 – Medium Impact

The system may still be operational, but prolonged high temperature can lead to:

• GPU thermal throttling

• Performance degradation

• Potential hardware damage if not resolved

Possible Causes

• High GPU workload or sustained compute jobs

• Cooling system inefficiency

• Fan malfunction or reduced airflow

• Data center ambient temperature increase

• Blocked air vents or dust accumulation

• Thermal throttling conditions

L1 Engineer Actions

Follow these steps when the alert is received.

Step 1 – Log the Alert

• Record the alert in the Alert Summary sheet

Step 2 – Create Ticket

• Open an incident ticket for tracking

Step 3 – Escalation (If Required)

If the issue appears significant:

• Create an internal ticket

• Inform L2 Engineer

Step 4 – Documentation

After confirmation with L2:

• Update the Issues Master Sheet

L2 Engineer Actions

Perform deeper analysis and corrective actions:

Step 1 - Detailed GPU and process check

Step 2 - Check airflow and cooling

Step 3 - Check thermal throttling

nvidia-smi -q | grep -i throttle

Step 4 - Analyze system workload

1. Identify abnormal or stuck GPU-intensive jobs.
2. Consider stopping or redistributing workloads if required.
   

Resolution

Depending on the root cause:

• Reduce GPU workload temporarily

• Improve airflow or cooling

• Replace faulty fans if detected

• Clean air vents or filters

• Adjust environmental cooling in the data center
  

• Server reboot / Hard reset

Escalation

Escalate to hardware support/vendor if:

• GPU temperature remains critical despite normal workload

• Cooling components are faulty

• GPU shows repeated thermal throttling