cuDNN
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. For more information go to https://developer.nvidia.com/cudnn.
To use cuDNN v5.1.10 for CUDA 7.5 you will need to add the following environment variables to your .bashrc file or wrapper script,
export CPATH=/usr/local/cudnn5.1-7.5/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cudnn5.1-7.5/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=/usr/local/cudnn5.1-7.5/lib64:$LIBRARY_PATH
Or load the environment module for cuDNN v5.1.10 for CUDA 7.5, by executing the following command,
module load cudnn/5.1-7.5
Note: This module also contains the correct CUDA paths.
cuDNN v5.1.10 for CUDA 7.5 documentation can be found in /usr/local/cudnn5.1-7.5/doc
cuDNN v5.1.10 for CUDA 7.5 sample can be found in /usr/local/cudnn5.1-7.5/sample
To use **cuDNN v6.0.21 for CUDA 7.5 **you will need to add the following environment variables to your .bashrc file or wrapper script,
export CPATH=/usr/local/cudnn6.0-7.5/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cudnn6.0-7.5/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=/usr/local/cudnn6.0-7.5/lib64:$LIBRARY_PATH
Or load the environment module for cuDNN v6.0.21 for CUDA 7.5, by executing the following command,
module load cudnn/6.0-7.5
Note: This module also contains the correct CUDA paths.
cuDNN v6.0.21 for CUDA 7.5 documentation can be found in /usr/local/cudnn6.0-7.5/doc
cuDNN v6.0.21 for CUDA 7.5 sample can be found in /usr/local/cudnn6.0-7.5/sample
To use cuDNN v5.1.10 for CUDA 8.0 you will need to add the following environment variables to your .bashrc file or wrapper script,
export CPATH=/usr/local/cudnn5.1-8.0/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cudnn5.1-8.0/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=/usr/local/cudnn5.1-8.0/lib64:$LIBRARY_PATH
Or load the environment module for v5.1.10 for CUDA 8.0, by executing the following command,
module load cudnn/5.1-8.0
Note: This module also contains the correct CUDA paths.
cuDNN v5.1.10 for CUDA 8.0 documentation can be found in /usr/local/cudnn5.1-8.0/doc
To use cuDNN v6.0.21 for CUDA 8.0 you will need to add the following environment variables to your .bashrc file or wrapper script,
export CPATH=/usr/local/cudnn6.0-8.0/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cudnn6.0-8.0/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=/usr/local/cudnn6.0-8.0/lib64:$LIBRARY_PATH
Or load the environment module for v6.0.21 for CUDA 8.0, by executing the following command,
module load cudnn/6.0-8.0
Note: This module also contains the correct CUDA paths.
To use cuDNN v7.0.5 for CUDA 8.0 you will need to add the following environment variables to your .bashrc file or wrapper script,
export CPATH=/usr/local/cudnn7.0-8.0/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cudnn7.0-8.0/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=/usr/local/cudnn7.0-8.0/lib64:$LIBRARY_PATH
Or load the environment module for v7.0.5 for CUDA 8.0, by executing the following command,
module load cudnn/7.0-8.0
Note: This module also contains the correct CUDA paths.
To use cuDNN v7.2.1 for CUDA 8.0 you need to add the following environment variables to your .bashrc file,
export CPATH=/usr/local/cudnn7.2-8.0/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cudnn7.2-8.0/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=/usr/local/cudnn7.2-8.0lib64:$LIBRARY_PATH
Or load the environment module for v7.2.1 for CUDA 8.0, by executing the following command,
module load cudnn/7.2-8.0
Note: This module also contains the correct CUDA paths.
To use cuDNN v7.0.5 for CUDA 9.0 you will need to add the following environment variables to your .bashrc file or wrapper script,
export CPATH=/usr/local/cudnn7.0-9.0/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cudnn7.0-9.0/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=/usr/local/cudnn7.0-9.0/lib64:$LIBRARY_PATH
Or load the environment module for v7.0.5 for CUDA 9.0, by executing the following command,
module load cudnn/7.0-9.0
Note: This module also contains the correct CUDA paths.
To use cuDNN v7.1.4 for CUDA 9.0 you will need to add the following environment variables to your .bashrc file or wrapper script,
export CPATH=/usr/local/cudnn7.1-9.0/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cudnn7.1-9.0/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=/usr/local/cudnn7.1-9.0/lib64:$LIBRARY_PATH
Or load the environment module for v7.1.4 for CUDA 9.0, by executing the following command,
module load cudnn/7.1-9.0
Note: This module also contains the correct CUDA paths.
To use cuDNN v7.2.1 for CUDA 9.0 you will need to add the following environment variables to your .bashrc file or wrapper script,
export CPATH=/usr/local/cudnn7.2-9.0/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cudnn7.2-9.0/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=/usr/local/cudnn7.2-9.0/lib64:$LIBRARY_PATH
Or load the environment module for v7.2.1 for CUDA 9.0, by executing the following command,
module load cudnn/7.2-9.0
Note: This module also contains the correct CUDA paths.
To use cuDNN v7.3.1 for CUDA 9.0 you will need to add the following environment variables to your .bashrc file or wrapper script,
export CPATH=/usr/local/cudnn7.3-9.0/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cudnn7.3-9.0/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=/usr/local/cudnn7.3-9.0/lib64:$LIBRARY_PATH
Or load the environment module for v7.3.1 for CUDA 9.0, by executing the following command,
module load cudnn/7.3-9.0
Note: This module also contains the correct CUDA paths.
To use cuDNN v7.4.2 for CUDA 9.0 you will need to add the following environment variables to your .bashrc file or wrapper script,
export CPATH=/usr/local/cudnn7.4-9.0/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cudnn7.4-9.0/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=/usr/local/cudnn7.4-9.0/lib64:$LIBRARY_PATH
Or load the environment module for v7.4.2 for CUDA 9.0, by executing the following command,
module load cudnn/7.4-9.0
Note: This module also contains the correct CUDA paths.
以此类推，下面我们只举这一个CUDA 10的一个例子
To use cuDNN v7.5.0 for CUDA 10.0 you will need to add the following environment variables to your .bashrc file or wrapper script,
export CPATH=/usr/local/cudnn7.5-10.0/include:$CPATH
export LD_LIBRARY_PATH=/usr/local/cudnn7.5-10.0/lib64:$LD_LIBRARY_PATH
export LIBRARY_PATH=/usr/local/cudnn7.5-10.0/lib64:$LIBRARY_PATH
Or load the environment module for v7.5.0 for CUDA 10.0, by executing the following command,
module load cudnn/7.5-10.0
Note: This module also contains the correct CUDA paths.
