Skip to content

Conversation

@camille-004
Copy link
Collaborator

Add basic GitHub Actions CI workflow to test runner infrastructure.

  • Tests Docker and NVIDIA CUDA image access
  • Includes GPU detection for RTX Pro 6000 runner
  • Follows NVIDIA GHA best practices
  • Ready for future source code integration

nvangelak and others added 6 commits June 13, 2025 17:07
- Simple CI workflow for testing runner infrastructure
- Tests Docker functionality and NVIDIA CUDA image access
- Uses CUDA 13.0.2 base image with Ubuntu 24.04
- Supports manual dispatch and runs on push/PR to main

Signed-off-by: cdunning <cdunning@nvidia.com>
- Add host GPU hardware detection with lspci and nvidia-smi
- Add container GPU access test with --gpus all flag
- Enable workflow triggers on cd/initial-ci branch for testing
- Will show RTX Pro 6000 GPU information when running on GPU runners

Signed-off-by: cdunning <cdunning@nvidia.com>
- Change runs-on from ubuntu-latest to linux-amd64-gpu-rtxpro6000-latest-1
- Re-enable branch trigger for cd/initial-ci to test GPU runner
- Should now run on the correct GPU-enabled self-hosted runner

Signed-off-by: cdunning <cdunning@nvidia.com>
Signed-off-by: cdunning <cdunning@nvidia.com>
- Consolidate 5 separate jobs into single matrix job
- Add conditional logic for GPU vs CPU testing
- Support both Linux and Windows platforms
- Enable/disable configurations via matrix.enabled flag
- Reduce code duplication by 90%

Signed-off-by: cdunning <cdunning@nvidia.com>
- Remove matrix.enabled condition (not available at job level)
- Split platform-specific steps with proper conditionals
- Separate GPU vs CPU testing steps
- Simplify matrix to only active configurations

Signed-off-by: cdunning <cdunning@nvidia.com>
- Remove shell condition from Test Docker step
- All steps now use default bash shell for Linux-only matrix

Signed-off-by: cdunning <cdunning@nvidia.com>
- Add GPU Windows x86_64 configuration
- Add CPU Windows x86_64 configuration
- Split CUDA image tests by OS (Linux vs Windows)
- Use PowerShell for Windows steps
- Now testing all 4 combinations: Linux/Windows × GPU/CPU

Signed-off-by: cdunning <cdunning@nvidia.com>
- GPU jobs: Use RTX Pro 6000 self-hosted runners
- CPU jobs: Use free GitHub runners (ubuntu-latest, windows-latest, ubuntu-24.04-arm64)
- Remove macOS configurations
- Optimize runner costs per NVIDIA best practices

Signed-off-by: cdunning <cdunning@nvidia.com>
- Use cmd shell instead of pwsh (not available)
- Remove Linux CUDA container tests on Windows
- Add simple Windows-native GPU/CPU tests
- Fix Windows batch syntax

Signed-off-by: cdunning <cdunning@nvidia.com>
- Add fallback GPU detection via wmic
- Handle nvidia-smi not being available on Windows runners
- Don't fail job if nvidia-smi missing

Signed-off-by: cdunning <cdunning@nvidia.com>
- Remove Windows GPU runner (no NVIDIA drivers installed)
- Remove ARM64 runner (not available on GitHub)
- Keep working configurations: GPU Linux + CPU Linux/Windows

Signed-off-by: cdunning <cdunning@nvidia.com>
- Use ubuntu-24.04-arm for Linux ARM64 builds
- Add architecture detection for ARM64
- Now supports x64 and ARM64 Linux builds

Signed-off-by: cdunning <cdunning@nvidia.com>
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

4 participants