Setting up NVIDIA GPU for deep learning: Installation of NVIDIA drivers, CUDA Toolkit and cuDNN in Ubuntu.

If one wants to train deep neural network models on largescale problems, GPUs are the way. Most of the existing deep learning libraries support both CPU and GPU. But, GPU based computations save a lot of your computation time. One can buy NVIDIA GPU from NVIDIA stores and place in PCI slot of your computers.

After placing GPU in your workstation, to run deep learning algorithm on GPU one has to install NVIDIA drivers, CUDA Toolkit and cuDNN. These three are essential to run deep learning tools on GPU.

Installation of NVIDIA Drivers:

Add the graphics-drivers ppa

sudo add-apt-repository ppa:graphics-drivers/ppa
sudo apt-get update

Purge any existing nvidia related packages you have installed

sudo apt purge nvidia*

Check which drivers are available for your system

ubuntu-drivers devices

Install the recommended driver

sudo apt-get install nvidia-390

Restart your system

sudo reboot

To know the whether drivers installed successfully or not, simply type nvidia-smi  in terminal.

nvidia-smi

It should show something like this....

CUDA Toolkit installation:

# Download cuda toolkit files from nvidia (.deb file) and root to the file directory
#You can download nvidia toolkit from here https://developer.nvidia.com/cuda-toolkit

# Root the .deb file directory (you can download latest version, the following lines for CUDA 9.1 version)

sudo dpkg -i cuda-repo-ubuntu1604-9-1-local_9.1.85-1_amd64.deb

sudo apt-get update

sudo apt-get install cuda

# Add the following lines to .bashrc

sudo gedit ~/.bashrc

# Paste the following lines at the end of the .bashrc file

export PATH=$PATH:/usr/local/cuda-9.1/binexport LD_LIBRARY_PATH=:/usr/local/cuda-9.1/lib64

# To know status, type nvcc --version in ternimal

nvcc --version

It should show something like this....

cuDNN installation:

#Download cuDNN file from Nvidia site
 (you can download latest release, the following lines for cudnn 9.1)

tar -zxf cudnn-9.1-linux-x64-v7.1.tgz

cd cuda

sudo cp lib64/* /usr/local/cuda/lib64/

sudo cp include/cudnn.h /usr/local/cuda/include/ 

Tensorflow installation

$pip install --upgrade tensorflow-gpu==1.4
$python
>>import tensorflow as tf