The NVIDIA CUDA Deep Neural Network library (cuDNN) is a GPU-accelerated library of primitives for deep neural networks. cuDNN provides highly tuned implementations for standard routines such as forward and backward convolution, pooling, normalization, and activation layers. cuDNN is part of the NVIDIA Deep Learning SDK.
Only GPU-enabled nodes are able to run the library. See Using GPUs on Iceberg for more information on how to request a GPU-enabled node for an interactive session or job submission.
Currently cuDNN 5.1 is available for CUDA versions 8.0.44 and 7.5.18. An appropriate CUDA module is loaded automatically so there’s no need for a separate CUDA module load call.
Load the appropriate cuDNN version with one of the following commands:
module load libs/binlibs/cudnn/5.1-cuda-8.0.44 module load libs/binlibs/cudnn/5.1-cuda-7.5.18
This section is primarily for administrators of the system.