cuDNN is a deep neural network library from Nvidia that provides a highly tuned implementation of many functions commonly used in deep machine learning applications. Exanples include forward and backword convolutions, activation layers, and normalization.

To view the available versions of cuDNN, run:

cuDNN contains both static and shared versions. If you need a different version installed, please contact the systems staff.

NVIDIA's page on cuDNN may be found on their developer Web site.