The problem
Getting CUDA workloads running on Windows via WSL2 involves a fragile chain of driver versions, kernel settings, and package conflicts. Most guides go out of date or miss critical steps.
Approach
A reproducible, step-by-step guide covering driver installation, WSL2 kernel configuration, CUDA Toolkit, cuDNN, and Docker GPU passthrough, version-pinned and maintained as a MkDocs documentation site.
Outcome
A living reference I use myself, and that the community references for Windows-based CUDA development. Covers the most common failure modes.
Highlights
- End-to-end: GPU driver → CUDA Toolkit → cuDNN → Docker GPU → Python.
- Version compatibility tables to avoid the most common mismatches.
- Covers Docker GPU passthrough for container-based ML workloads.
- Built and maintained with MkDocs.