In this tutorial I will show you how to install tensorflow. After spending a week and testing every available option to install the tensorflow platform, I realized that none are complete or time less (in the sense that after the new version is released, you cannot use them and they are considered outdated). also some methods use not uptodate libraries. I used CUDA 9.2 with CuDNN 7.2 along NCCL 2.2 and tensorflow 1.10, but because I downloaded everything from their webpages, should be ok for later releases.

I have tried to install tensorflow in a python Virtual Environment, but it seems that I cannot access it in there, but I have it outside of it anyways!!

#Packages You are going to need some packages that need installation before the actual tensorflow: (I assume the system is brand new)

$ sudo apt-get update
$ sudo apt-get upgrade
$ sudo apt-get install python2.7-dev python3.5-dev

As mentioned it virtualization did not work for me, but I mention it anyways:

$ cd ~
$ wget https://bootstrap.pypa.io/get-pip.py
$ sudo python get-pip.py
$ sudo pip install virtualenv virtualenvwrapper
$ sudo rm -rf ~/get-pip.py ~/.cache/pip
$ echo -e "\n# virtualenv and virtualenvwrapper" >> ~/.bashrc
$ echo "export WORKON_HOME=$HOME/.virtualenvs" >> ~/.bashrc
$ echo "source /usr/local/bin/virtualenvwrapper.sh" >> ~/.bashrc
$ source ~/.bashrc

Use python3 for another version:

$ mkvirtualenv cv -p python2

Now rest of the packages

$ pip install numpy

Use this for python2:

$ sudo apt-get install python-numpy python-dev python-pip python-wheel

Use this for python3:

$ sudo apt-get install python3-numpy python3-dev python3-pip python3-wheel

I did not see the following anywhere, but I received errors for them, so I mention them: use pip3 for python3:

Pip install enum34
Pip install mock

###Nvidia Driver Look for the “***” version of the driver on Nvidia website

$ sudo add-apt-repository ppa:graphics-drivers/ppa
$ sudo apt-get update
$ sudo apt-get install nvidia-*** nvidia-modprobe

Reboot your system and run “nvidia-smi” and you should see a table with the version of the driver at top.

There are many methods to install CUDA, but I prefer to download the file from Nvidia and install it, to have the most updated version! Go to https://developer.nvidia.com/cuda-zone ->download now, Linux, x86_64, Ubuntu, 16.8 (or later), deb(local) Go to download directory and then do:

$ sudo dpkg -i cuda-repo-ubuntu1604-9-2-local_9.2.148-1_amd64.deb

If gives an error about key, just read it, tells you what to put and then run “dpkg” again.

$ sudo apt-key add /var/cuda-repo-<version>/7fa2af80.pub
$ sudo apt-get update
$ sudo apt-get install cuda

Even if “nvcc –version” does not work, it should be fine now and you have CUDA if everything goes ok.

I have installed it using the .deb file and during tensorflow compile got an error and did it using the .tar and could not test to see if it is ok, so I did both and everyone was happy!!! Go to https://developer.nvidia.com/cudnn and using your account download the files for your CUDA. You should download : cuDNN v*** Library for Linux cuDNN v*** Runtime Library for Ubuntu16.04 (Deb) cuDNN v*** Developer Library for Ubuntu16.04 (Deb) cuDNN v*** Code Samples and User Guide for Ubuntu16.04 (Deb) then go to download directory and do (based on the CUDA-**, the version of CUDA you installed) my version was cuda-9.2:

$ tar -xzvf cudnn-**-linux-x64-v7.tgz
$ sudo cp cuda/include/cudnn.h /usr/local/cuda-**/include
$ sudo cp cuda/lib64/libcudnn* /usr/local/cuda-**/lib64
$ sudo chmod a+r /usr/local/cuda-**/include/cudnn.h /usr/local/cuda-**/lib64/libcudnn*

Then:

$ sudo dpkg -i libcudnn7_***-1+cuda**_amd64.deb
$ sudo dpkg -i libcudnn7-dev_***-1+cuda**_amd64.deb
$ sudo dpkg -i libcudnn7-doc_***-1+cuda**_amd64.deb

Then test the CuDNN to make sure:

$cp -r /usr/src/cudnn_samples_v7/ $HOME
$ cd  $HOME/cudnn_samples_v7/mnistCUDNN
$make clean && make
$ ./mnistCUDNN

If there is a library issue, make sure that you add your lib64 to the PATH:

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64

or look at the following first answer for the permanent change: https://stackoverflow.com/questions/13428910/how-to-set-the-environmental-variable-ld-library-path-in-linux

If you see: “Test passed!” at the end you have it up and going. TensorRT and NCCL are also available for installation and their documentation is valid in Nvidia website.

I found the following for it, should be a simple google search in anycase:

$ sudo apt-get install openjdk-8-jdk
$ echo "deb [arch=amd64] http://storage.googleapis.com/bazel-apt stable jdk1.8" | sudo tee /etc/apt/sources.list.d/bazel.list
$ curl https://bazel.build/bazel-release.pub.gpg | sudo apt-key add -
$ sudo apt-get update
$ sudo apt-get install bazel

If everything is good so far, you should be ok to install tensorflow with GPU support now:

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/extras/CUPTI/lib64
$ sudo apt-get install git
$ git clone https://github.com/tensorflow/tensorflow
$ cd tensorflow
$ git checkout r1.**

You can put the version you want to install here, which changes with the new release, just look at their website for the latest.

$ cd tensorflow
$ ./configure

Asks you for information before compile, The order and number of questions may change, but they should be pretty forward to answer, if you don’t know an answer, press ENTER for default. My 1.10 version asked for (your answers may differ):

Python path

Python libraries path

Jemalloc support :yes

Google cloud support: no

Hadoop support: no

Amazon cloud support: no

XLA support: no

VERBS support: no

OpenCL support: no

CUDA support: yes

CUDA version: 9.2

CUDA path: /usr/local/cuda-9.2

Cudnn version: 7.2

Cudnn path: default

tensorRT support: no

NCCL:1.3

Android support: no

$ bazel build --config=opt --config=cuda //tensorflow/tools/pip_package:build_pip_package
$ bazel-bin/tensorflow/tools/pip_package/build_pip_package /tmp/tensorflow_pkg
$ sudo pip install /tmp/tensorflow_pkg/tensorflow-***-py2-none-any.whl

$ cd ~
$ python
>>> import tensorflow as tf
>>> sess = tf.Session(config=tf.ConfigProto(log_device_placement=True))

It should give you a hardware list and summery, (feels nice at this point to read that summery)