/benchmark-gpu

Run comprehensive GPU benchmarks and document GPU module configuration:

1. GPU Detection:

   • List GPUs: lspci | grep -i vga
   • GPU details: lspci -v -s $(lspci | grep -i vga | cut -d' ' -f1)

2. NVIDIA GPU (if present):

   • nvidia-smi output: nvidia-smi
   • Driver version: nvidia-smi --query-gpu=driver_version --format=csv,noheader
   • GPU details: nvidia-smi --query-gpu=name,memory.total,compute_cap --format=csv
   • CUDA version: nvcc --version or nvidia-smi | grep "CUDA Version"
   • CUDA libraries: ldconfig -p | grep cuda

3. AMD GPU (if present):

   • ROCm info: rocm-smi or /opt/rocm/bin/rocminfo
   • ROCm version: apt list --installed | grep rocm or check /opt/rocm/.info/version
   • GPU details: rocm-smi --showproductname --showvbios
   • HIP version: hipconfig --version (if available)

4. GPU Performance Testing:

   • NVIDIA: nvidia-smi dmon -c 10 (10-second monitoring)
   • AMD: rocm-smi -d (device info)
   • Check GPU memory bandwidth if tools available
   • Temperature monitoring: Check max temps under idle

5. OpenCL/Vulkan:

   • OpenCL devices: clinfo (if installed)
   • Vulkan support: vulkaninfo --summary (if installed)

6. GPU Compute Benchmarks (if tools available):

   • For NVIDIA: Consider glmark2, unigine, or pytorch GPU tests
   • For AMD: Consider ROCm benchmark utilities
   • Document if benchmark tools aren't available

7. Passthrough Status (for VMs):

   • Check IOMMU groups: find /sys/kernel/iommu_groups/ -type l
   • Verify GPU in VM: lspci -nnk | grep -A3 VGA

Create a comprehensive GPU profile document and save it to hw-profile/gpu-benchmark-$(date +%Y%m%d-%H%M%S).md with:

• GPU model and specifications
• Driver versions (NVIDIA/AMD)
• CUDA/ROCm installation status and versions
• Compute capability
• Available GPU libraries
• Performance metrics
• Temperature and power data
• Passthrough configuration (if VM)
• OpenCL/Vulkan support
• Benchmark results (if run)
• Timestamp

Include recommendations for:

• Driver updates if needed
• CUDA/ROCm optimization
• Cooling concerns if high temps
• Compute capability utilization

Format with clear sections and code blocks for raw outputs.

After creating the document, inform Daniel of the location and highlight any configuration issues or optimization opportunities.