Roadmap

Track development progress as features are implemented.

MVP (v0.1)

• Install CUDA + cuDNN together (cudup install)
• List available/installed versions (cudup list)
• Switch between versions (cudup use)
• Shell integration (cudup manage setup) (bash, zsh, fish)
• Health diagnostics (cudup check)
• Basic error handling and user-friendly messages
• Linux support (Ubuntu/Debian)

v0.5

• Per-project .cuda-version files (cudup local)
• Show current version (cudup current)
• Automatic cuDNN version matching
• Uninstall versions (cudup uninstall)
• Clean up old versions (cudup clean)
• Diagnostics (runtime checks, GPU detection)
• Show cudup version (cudup --version)
• Checksum verification (SHA256 integrity checks)
• Resumable downloads for large installers
• Pre-flight compatibility checking (GPU driver, compute capability)

v1.0

• NCCL support (--with-nccl)
• TensorRT support (--with-tensorrt)
• Framework compatibility checking (cudup check pytorch/tensorflow)
• Configuration file support (~/.cudup/config.toml)
• Self-update mechanism (cudup manage self-update)
• Remote version manifest (fetch latest available versions)
• Proxy configuration for corporate environments
• PATH rollback support (cudup manage remove --rollback)
• Advanced installation options (--minimal, --from-cache)
• Global vs local version modes (cudup global)
• Import existing installations (cudup import)
• Export/import environments
• Multi-architecture support (x86_64, ARM64/SBSA)
• Multi-distro support (RHEL, CentOS, Fedora)
• Performance optimizations (parallel downloads)
• Comprehensive documentation
• Automated testing suite

Future Considerations

• WSL2 support
• Native Windows support (if demand exists?)
• Docker integration
• Plugin system for extensions