Analysis

Compare the findings from the previous steps.

Scenario 1 – Active Hardware Issue

Indicators include:

• NVSM reports unhealthy hardware.
• CPU-related entries exist in the BMC SEL.
• Linux kernel reports Machine Check Exceptions.
• The alert reoccurs after clearing.

In this case:

• Treat the alert as a hardware issue.
• Collect diagnostics.
• Open a support case with NVIDIA.

Scenario 2 – Historical or Stale Alert

Indicators include:

• System health is healthy.
• No CPU-related SEL entries.
• No Machine Check Exceptions in the operating system logs.
• The alert exists only in the NVSM alert database.

In this case, the alert may be historical and can typically be cleared after verifying that no active hardware fault exists.

Clearing Historical Alerts

If NVIDIA Support confirms that the alert is stale, clear the NVSM database and BMC SEL.

Stop NVSM: sudo systemctl stop nvsm

Remove the NVSM alert database: sudo rm /var/lib/nvsm/sqlite/nvsm.db

Clear the BMC SEL:  sudo ipmitool sel clear

Remove archived SEL logs: sudo rm /var/log/bmc_sel_archive_for_BMC_*.log

Restart NVSM: sudo systemctl start nvsm

Verification

After cleanup, verify:

sudo nvsm show health
sudo nvsm show alerts
sudo ipmitool sel list

Expected results:

• Overall system health is healthy.
• No active NV-CPU-XX alert.
• No CPU-related hardware events in the SEL.

When to Open a Support Case

Contact NVIDIA Enterprise Support if:

• The alert persists after verification and cleanup.
• The alert reappears after being cleared.
• CPU-related SEL entries are present.
• Machine Check Exceptions are reported.
• The system repeatedly enters degraded mode.
• CPU hardware is reported as unhealthy.

When opening a case, include:

• NVSM health dump (sudo nvsm dump health)
• BMC SEL output (ipmitool sel list)
• Kernel logs (journalctl -k)
• Relevant operating system logs