Title: HP ProLiant SL390s G7 Server - 4U: How to Locate a Specific GPU in Server?
Object Name: emr_na-kc0114246en_us
Document Type: Support Information
Original owner: KCS - ProLiant Servers
Disclosure level: Public
Version state: final
Environment
FACT:Proliant SL390s G7 4U
FACT:Multiple Nvidia GPUs installed in system
FACT:Linux operating system
Questions/Symptoms
SYMPTOM:System logs show error on a specific GPU as in following example:
Jan 24 22:02:23 servername kernel: NVRM: GPU at 0000:0a:00: GPU-344da356-e1bb-8265-f314-9c57c2bac794
Jan 24 22:02:23 servername kernel: NVRM: Xid (0000:0a:00): 48, An uncorrectable double bit error (DBE) has been detected on GPU (00 03 00).
GOAL:Identify defective GPU across all GPUs so it can be replaced.
Cause
CAUSE:Defective GPU
Answer/Solution
FIX:Use bus address (0a in above example) and following table to identify GPU physical slot:
|
| SL390s_4U_GPU_mapping.jpg |
FIX:"nvidia-smi -q" command also lists GPUs UUID and serial numbers and might be used to validate GPU to be replaced by looking at GPU label. "nvidia-smi -q" output is already included as result of "/usr/bin/nvidia-bug-report.sh" script.© Copyright 2025 Hewlett Packard Enterprise Development Company, L.P.
Hewlett Packard Enterprise believes in being unconditionally inclusive. Efforts to replace noninclusive terms in our active products are ongoing.