GPU & Guest Tools Troubleshooting
Validate the GPU path in layers: PCI device, Guest Tools driver and agent, control device, STRATUM inventory, CUDA/NVML validation, then the application.
Verify in layers
Confirm the PCI device first, then the STRATUM guest driver/control nodes, then Guest Tools agent state, then STRATUM GPU inventory, then CUDA/NVML validation, then the application. Stop at the first layer that disagrees with the expected state.
Keep normal setup normal
Do not add manual LD_LIBRARY_PATH overrides, native-provider ports, or mock-mode variables to ordinary customer troubleshooting. Those were development plumbing, not the intended product configuration path.
Commands
Run these checks inside the guest. They verify the supported customer path and intentionally omit development-only library-path, mock-mode, and native-port overrides.
lspci -nn | grep -i -E '3d|display|1d1e:5350'
lsmod | grep stratum_gpu
ls -l /dev/stratum-gpu*
ls -l /dev/stratum-gpuctl
systemctl status stratum-guest-agent
stratum-gpu-query
stratum-gpu-smi