git.png

    Install CUDA & cuDNN:

    If you want to use the GPU version of the TensorFlow you must have a cuda-enabled GPU.

    • To check if your GPU is CUDA-enabled, try to find its name in the long list of CUDA-enabled GPUs.
    • To verify you have a CUDA-capable GPU:
      • (for Windows) Open the command prompt (click start and write “cmd” on search bar) and type the following command: bash control /name Microsoft.DeviceManager
      • (for Linux) Open terminal (Alt+Ctrl+T) and type: bash lspci | grep -i nvidia

    You need to install CUDA and cuDNN with following versions:

    • CUDA tooklit: 9.0
    • cuDNN: 7.0.5

    Windows:

    1. Download and install the CUDA toolkit 9.0 from https://developer.nvidia.com/cuda-90-download-archive. Choose the correct version of your windows and select local installer:

    Alt text

    Install the toolkit from downloaded .exe file.

    2. Download the cuDNN v7.0.5 (CUDA for Deep Neural Networks) library from here. It will ask for setting up an account … (it is free) Download cuDNN v7.0.5 for CUDA 9.0.

    Choose the correct version of your Windows. Download the file. Copy the files to “C:\Program FIles\NVIDIA GPU Computing Toolkit\CUDA\v9.0” in the corresponding folders:

    Alt text

    Linux:

    1. Download and install the CUDA toolkit 9.0 from https://developer.nvidia.com/cuda-90-download-archive. Choose the correct version of your Linux and select runfile (local) local installer: Alt text

    Open terminal (Alt+Ctrl+T) and type:

    chmod +x cuda_9.0.176_384.81_linux.run
    sudo ./cuda_9.0.176_384.81_linux.run --override
     

    *Note: Do not install the Graphics Driver.

    You can verify the installation:

    nvidia-smi
     

    Also you can check where your cuda installation path (we will call it as <cuda_path>) is using one of the commands:

    which nvcc
    ldconfig -p | grep cuda
     

    Your <cuda_path> will be /usr/... or /usr/local/cuda/ or /usr/local/cuda/cuda-9.0/. Locate it and add it to your .bashrc file:

    export CUDA_ROOT=<cuda_path>/bin/
    export LD_LIBRARY_PATH=<cuda_path>/lib64/
     

    2. Download the cuDNN v7.0.5 (CUDA for Deep Neural Networks) library from here. It will ask for setting up an account … (it is free) Download cuDNN v7.0.5 for CUDA 9.0.

    Choose cuDNN v7.0.5 Library for Linux. Go to the folder that you downloaded the file and open terminal (Alt+Ctrl+T):

    tar -xvzf cudnn-9.0-linux-x64-v7.tgz
    cd cuda/
    sudo cp -P include/cudnn.h <cuda_path>/include
    sudo cp -P lib64/libcudnn* <cuda_path>/lib64
    sudo chmod a+r <cuda_path>/lib64/libcudnn*
     

    3. Install libcupti-dev: bash sudo apt-get install libcupti-dev

    © 2018 Easy-TensorFlow team. All Rights Reserved.