cudnn-info
Detect and validate cuDNN library installations.
Usage
What It Shows
cuDNN Detection
- cuDNN version and build information
- Library file locations
- Header file locations
Installation Validation
- Multiple installation detection
- Symlink validation (Linux)
- PATH configuration (Windows)
Compatibility
- CUDA version compatibility
- Version requirements for popular frameworks
Example Output
🧠 cuDNN ANALYSIS
============================================================
📚 cuDNN Installation:
Version: 8.9.7
CUDA Version: 12.x
Libraries:
✅ libcudnn.so.8.9.7
└─ Path: /usr/local/cuda-12.1/lib64/libcudnn.so.8
Headers:
✅ cudnn.h
└─ Path: /usr/local/cuda-12.1/include/cudnn.h
🔗 Symlinks (Linux):
libcudnn.so → libcudnn.so.8 → libcudnn.so.8.9.7
✅ Symlink chain is valid
🎮 Compatibility:
✅ cuDNN 8.9.7 is compatible with CUDA 12.1
✅ Meets PyTorch 2.1+ requirements (cuDNN 8.5+)
✅ Meets TensorFlow 2.15+ requirements (cuDNN 8.6+)
Multiple Installations
⚠️ Multiple cuDNN installations detected
/usr/local/cuda-12.1/lib64/libcudnn.so.8.9.7
/usr/lib/x86_64-linux-gnu/libcudnn.so.8.6.0
💡 Recommendations:
- Ensure LD_LIBRARY_PATH prioritizes the correct version
- Consider removing older installations to avoid conflicts
Common Issues
Missing cuDNN
❌ cuDNN not found
💡 Installation:
1. Download from https://developer.nvidia.com/cudnn
2. Extract to /usr/local/cuda-12.1/
3. Or install via package manager:
sudo apt install libcudnn8 libcudnn8-dev
Broken Symlinks
❌ Broken symlink detected
libcudnn.so → libcudnn.so.8 (missing)
💡 Fix:
cd /usr/local/cuda/lib64
sudo ln -sf libcudnn.so.8.9.7 libcudnn.so.8
sudo ln -sf libcudnn.so.8 libcudnn.so
Version Mismatch
⚠️ cuDNN version may not be optimal
Installed: cuDNN 8.2.0
PyTorch 2.1 recommends: cuDNN 8.5+
💡 Consider upgrading cuDNN for better performance