Verifying CuDNN Installation

CuDNN (CUDA Deep Neural Network library) is a GPU-accelerated library of primitives for deep neural networks. Verifying its installation is crucial to ensure that your system can leverage the power of NVIDIA GPUs for deep learning computations. This tutorial will guide you through the steps to verify CuDNN installation on Linux and Windows systems.

Understanding CuDNN Installation

CuDNN installation involves copying specific files into your CUDA directory. The verification process, therefore, focuses on checking the presence and version of these files.

Verifying CuDNN Installation on Linux

1. Checking CuDNN Version:
   To verify if CuDNN is installed and to check its version, you can use the whereis command to locate the cudnn.h file, followed by a grep command to extract version information.

   cat $(whereis cudnn.h) | grep CUDNN_MAJOR -A 2
   

   If this doesn’t work, try:

   cat $(whereis cuda)/include/cudnn.h | grep CUDNN_MAJOR -A 2
   

   Alternatively, if you know the path to your CUDA installation (commonly /usr/local/cuda/), you can directly check the version:

   cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2
   

   Or for newer versions where cudnn_version.h is used:

   cat /usr/local/cuda/include/cudnn_version.h | grep CUDNN_MAJOR -A 2
   

2. Checking CuDNN Library:
   Another way to verify the installation is by checking if the CuDNN library (libcudnn.so) exists in your CUDA lib directory.

   ls /usr/local/cuda/lib64/libcudnn*
   

   If the file is present, it indicates a successful installation.

3. Using ldconfig and grep:
   You can also use ldconfig with grep to verify if CuDNN is recognized by your system.

   /sbin/ldconfig -N -v $(sed 's/:/ /' <<< $LD_LIBRARY_PATH) 2>/dev/null | grep libcudnn
   

Verifying CuDNN Installation on Windows

1. Checking CuDNN Version:
   First, locate the path to your cudnn.h file using the where command in the Command Prompt.

   C:\>where cudnn*
   

   This will output paths where CuDNN files are found. Look for the cudnn.h file location.

   Then, use the type command to view the contents of cudnn.h and find the version information:

   type "%PROGRAMFILES%\cuDNN6\cuda\include\cudnn.h" | findstr "CUDNN_MAJOR CUDNN_MINOR CUDNN_PATCHLEVEL"
   

2. Checking CUDA Version:
   On both Linux and Windows, you can check the CUDA version (which indirectly verifies that your system is set up for GPU acceleration) using:

   nvcc --version
   

   or on Windows:

   C:\>nvcc --version
   

Conclusion

Verifying CuDNN installation involves checking for the presence and version of specific files within your CUDA directory. By following these steps, you can ensure that your system is properly set up to utilize CuDNN for deep learning tasks.

Troubleshooting Tips

• If cudnn.h or libcudnn.so are not found, double-check your installation path and ensure that the files were correctly copied during installation.
• For issues related to version compatibility (e.g., errors with TensorFlow), consider reinstalling CuDNN with a version known to be compatible.

By mastering these verification techniques, you’ll be well-equipped to diagnose and resolve common issues related to CuDNN installations on both Linux and Windows platforms.