Tensorflow installation on Ubuntu

Tensorrflow installation on Ubuntu
----------------------------------------

Required Steps:
------------------
  - Install Nvidia GPU drivers
  - Install Cuda
  - Install CudNN
  - Install Anaconda
  - Install tensorflow
  - Run sample program to test


Install Nvidia GPU drivers
---------------------------

lspci | grep NVIDIA  // to see if graphic card is connected to the machine


Add the drivers repository using the following command
sudo add-apt-repository ppa:graphics-drivers/ppa

Update and install NVIDIA drivers using the following two commands
sudo apt update
sudo apt install nvidia-390

Restart the machine

After restart type in the following command to verify if graphic drivers are installed
nvidia-smi


-- NVIDIA DRIVERS ARE INSTALLED --



Install Cuda
--------------

As Tensorflow currently support CUDA 9.0 only we are going to install CUDA 9.0 thought the latest version of CUDA is 10.0


Download appropriate CUDA distribution run file from the following link suitable for your operaing system
https://developer.nvidia.com/cuda-90-download-archive


chmod +x cuda_9.0.176_384.81_linux.run

to install CUDA, run the .run file using the following command and follow the steps in the wizard
./cuda_9.0.176_384.81_linux.run

After the installation is complete, add the following lines to ~/.bashrc at the end

export CUDA_HOME=/usr/local/cuda
export PATH=$CUDA_HOME/bin:$PATH
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64
export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/extras/CUPTI/lib64


Open /etc/ld.so.conf and add the following line at the end
/usr/local/cuda-9.0/lib64

run the following command
sudo ldconfig


Once this command runs successfully, verify CUDA installation using the following command
 nvcc --version

This command should print out CUDA version

Now enable persistence mode for the GPU using the following command
sudo nvidia-smi -pm 1



-- CUDA is Installed --




Install CudNN
--------------

As we are using CUDA 9.0 download CUDNN version that matches CUDA-9.0 and download a generic Linux distribution from the following link
https://developer.nvidia.com/rdp/cudnn-download

Extract and copy the files to respective locations using the following commands

tar -xvzf cudnn-9.0-linux-x64-v7.4.2.24.tgz
sudo cp cuda/include/cudnn.h /usr/local/cuda/include/
sudo cp cuda/lib64/* /usr/local/cuda/lib64/
sudo chmod a+r /usr/local/cuda/lib64/libcudnn*


-- CUDNN is installed --



Install Anaconda
----------------

Download Anaconda Python version that Tensorflow supports, here we have downloaded Python 3.6 from the following link
https://repo.anaconda.com/archive/

chmod +x Anaconda3-5.1.0-Linux-x86_64 

Install Anaconda Python using the following command and follow the wizard
./Anaconda3-5.1.0-Linux-x86_64

Once the installation is done, open a new terminal and verify python installation using
python --version


-- Anaconda Python is installed --



Install Tensorflow
------------------

Install Tensorflow using the following command
pip install tensorflow-gpu

-- Tensorflow is installed --

Sample program to Test
--------------------------
Once the installation completes, verify the installation using the following command
python -c "import tensorflow as tf; tf.enable_eager_execution(); print(tf.reduce_sum(tf.random_normal([1000, 1000])))"

From the logs you can see that Tensorflow using GPU to compute