docs
docs copied to clipboard
Fix outdated CUDA version and add GPU setup note for Model Builder
This PR updates the Model Builder GPU installation guide to resolve issues reported when using CUDA 10.1, which caused "no kernel found" errors during object detection while image classification worked correctly.
Changes made
Updated the recommended CUDA version from 10.1 → 11.8 (or later).
Added a clear warning note about kernel errors with older CUDA versions.
Updated cuDNN installation instructions for compatibility with CUDA 11.x.
Retained and clarified the troubleshooting section for driver setup and CUDA detection.
Updated metadata date and improved formatting for clarity.
Why this change
Following the previous instructions (CUDA 10.1 + cuDNN 7.6.4) led to TensorFlow GPU initialization issues for object detection models in Model Builder. The updated instructions now reflect the tested, stable configuration for recent .NET and TensorFlow versions.
Internal previews
| 📄 File | 🔗 Preview link |
|---|---|
| docs/core/tools/dotnet-run.md | dotnet run |
| docs/machine-learning/how-to-guides/install-gpu-model-builder.md | How to install GPU support in Model Builder |