This document describes how to install the combination of 14.04 + CUDA 7.5 + Tensorflow. This combination is the easiest to install without anything like compilation from sources etc.

Download and install Ubuntu 14.04 or 16.04

Install CUDA 8

This way below installs CUDA and all the related things (e.g. drivers needed).

Go to https://developer.nvidia.com/cuda-downloads
Choose Linux -> x86_64 -> Ubuntu -> 14.04 or 16.04 -> deb (local)
Press download

After download is over, open a terminal and navigate to Downloads.

wget https://s3.amazonaws.com/workspace.insightzen.com/bins/cuda-repo-ubuntu1604-8-0-local-ga2_8.0.61-1_amd64.deb
sudo dpkg -i cuda-repo-ubuntu1604-8-0-local-ga2_8.0.61-1_amd64.deb
sudo apt-get update
sudo apt-get install cuda

the dpkg -i cuda-repo-ubuntu1404-8-0-local-ga2_8.0.61-1_amd64.deb might differ (check the download name - or use tab for autocomplete)

Restart the PC to activate CUDA + the new drivers.

Install CUDNN 5.1

CUDNN makes CNNs faster with some convolutional and other optimizations.

Navigate to https://developer.nvidia.com/cudnn and register for an account (it's free). After you make your account, login and go to downloads.

Choose the following Download cuDNN v5.1 (August 10, 2016), for CUDA 8 cuDNN v5.1 Library for Linux

after it is downloaded navigate to Downloads and extract the tar file. You will get a folder called cuda. Open a terminal and run the following to navigate to this folder and put the cudnn files to your system folders.

wget https://s3.amazonaws.com/workspace.insightzen.com/bins/cudnn-8.0-linux-x64-v5.1.tgz
tar xvzf cudnn-8.0-linux-x64-v5.1.tgz
cd cuda
sudo cp lib64/* /usr/local/cuda/lib64/
sudo cp include/cudnn.h /usr/local/cuda/include/

Update your .bashrc

Open .bashrc from your home folder with an editor (the dot indicates that it is a hidden file). a way to do this for example is

vim ~/.bashrc

Add the following lines to the end of the file.

# add cuda tools to command path
export PATH=/usr/local/cuda/bin:${PATH}
export MANPATH=/usr/local/cuda/man:${MANPATH}

# add cuda libraries to library path
if [[ "${LD_LIBRARY_PATH}" != "" ]]
then
  export LD_LIBRARY_PATH=/usr/local/cuda/lib64:${LD_LIBRARY_PATH}
else
  export LD_LIBRARY_PATH=/usr/local/cuda/lib64
fi

You need to log out and log back in after this step.

Install tensorflow

wget https://repo.continuum.io/archive/Anaconda3-4.4.0-Linux-x86_64.sh
bash Anaconda3-4.4.0-Linux-x86_64.sh
conda create -n tensorflow
source activate tensorflow
pip install --ignore-installed --upgrade tfBinaryURL

where tfBinaryURL is the URL of the TensorFlow Python package. For example, the following command installs the GPU version of TensorFlow for Python 3.6:

(tensorflow)$ pip install --ignore-installed --upgrade \
 https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.2.0-cp36-cp36m-linux_x86_64.whl

Enjoy your new gpu-enabled deep learning setup :)

For windows

https://ulrik.is/writing/cuda-8-0-cudnn-5-tensorflow-1-0-and-keras-on-windows-10/

Reference :

RooseveltAdvisors/tensorflow-cuda-ubuntu.md