nvidia-bug-report-gpu-operator
Installation
SKILL.md
nvidia-bug-report on GPU Operator–Managed Nodes
Overview
On GPU Operator–managed nodes the NVIDIA driver runs inside a container and is bind-mounted onto the host at /run/nvidia/driver. Consequences:
nvidia-smi,nvidia-bug-report.sh, etc. are not on the host$PATHdpkg -l | grep nvidiashows nothing on the host — that's normal, not a broken install- Do not
apt install nvidia-utils-*— it installs a second, conflicting driver stack on top of the Operator-managed one - The kernel modules (
nvidia,nvidia_uvm,nvidia_modeset) ARE loaded into the host kernel and show up inlsmod
Locate the tools
find / -iname "nvidia-bug-report*" 2>/dev/null
# typically: /run/nvidia/driver/usr/bin/nvidia-bug-report.sh