How to install tensorflow-gpu without shitting on your computer ?

Once upon a time

There was a operating system called Ubuntu 14.04, I tried installing tensorflow gpu on that. And I was stuck in login loop as Tom cruise was stuck in time loop in the movie “The Edge of tomorrow”. I couldn’t get out of login loop until I installed Ubuntu 16.04

Want to know how was my experience with ubuntu 16.04

In a very few[sarcastic laugh] steps I was able to install it.

1. sudo apt-get –purge remove nvidia-*
2. sudo apt-get autoremove
3. sudo apt-get update
4. sudo gedit /etc/modprobe.d/blacklist-nouveau.conf
   paste

   blacklist nouveau
   blacklist lbm-nouveau
   options nouveau modeset=0
   alias nouveau off
   alias lbm-nouveau off
   

5. echo options nouveau modeset=0 | sudo tee -a /etc/modprobe.d/nouveau-kms.conf
6. sudo update-initramfs -u
7. sudo apt-get update
8. sudo apt-get upgrade -y
9. sudo apt-get dist-upgrade -y
10. sudo apt-get install build-essential
11. sudo apt-get install linux-image-extra-virtual
12. sudo apt-get install linux-source
13. sudo apt-get source linux-image-$(uname -r)
14. sudo apt-get install linux-headers-$(uname -r)
15. wget -O cuda_8_linux.run https://developer.nvidia.com/compute/cuda/8.0/Prod2/local_installers/cuda_8.0.61_375.26_linux-run
16. sudo chmod +x cuda_8_linux.run

    #in tty ctrl+alt+f1 or ctrl + alt +f6
    sudo service lightdm stop 
    #or if you are using gnome 3 then
    sudo service gdm3 stop
    sudo ./cuda_8_linux.run
    

17. sudo service gdm3 stop
18. sudo ./cuda_8_linux.run

    #Do you accept the previously read EULA?
    #accept
    #Install NVIDIA Accelerated Graphics Driver for Linux-x86_64 367.48?
    #n (we installed drivers previously)
    #Install the CUDA 8.0 Toolkit?
    #y
    #Enter Toolkit Location:
    # /usr/local/cuda-8.0 (enter)
    # Do you wish to run the installation with ‚sudo’?
    # y
    # Do you want to install a symbolic link at /usr/local/cuda?
    # y 
    # Install the CUDA 8.0 Samples?
    # y 
    # Enter CUDA Samples Location:
    # enter 
    # Install cuDNN
    # go to website and download cudnn-8.1 https://developer.nvidia.com/cudnn
    

19. tar -zxvf cudnn-8.1-linux-x64-v5.1.tgz

20. copy libs to /usr/local/cuda folder
21. sudo cp -P cuda/include/cudnn.h /usr/local/cuda/include
22. sudo cp -P cuda/lib64/libcudnn* /usr/local/cuda/lib64
23. sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn*
24. sudo add-apt-repository ppa:graphics-drivers/ppa
25. sudo apt-get update
26. sudo apt-get install nvidia-375
27. sudo apt-get install libcupti-dev

    Once nvidia driver is installed, restart the computer. You can verify the driver using the following command.

28. cat /proc/driver/nvidia/version
29. echo ‘export PATH=/usr/local/cuda/bin:$PATH’ » ~/.bashrc
30. sudo apt-get install python-numpy python-dev python-pip python-wheel

    Download Bazel

31. chmod +x bazel-0.5.2-installer-linux-x86_64.sh
32. ./bazel-0.5.2-installer-linux-x86_64.sh –user

    paste in bashrc

33. export PATH=”$PATH:$HOME/bin”

    https://www.anaconda.com/download/ install anaconda

    Add these two lines in gedit ~/.bashrc

34. export LD_LIBRARY_PATH=”$LD_LIBRARY_PATH:/usr/local/cuda/lib64”
35. export CUDA_HOME=/usr/local/cuda
36. conda create -n tensorflow
37. source activate tensorflow
38. pip install –ignore-installed –upgrade \ https://storage.googleapis.com/tensorflow/linux/gpu/tensorflow_gpu-1.0.0-cp27-none-linux_x86_64.whl

But as time has passed by…………..

Things have changed a lot.Now on new ubuntu 18.04 you don’t need to do all this, Things have become relatively easy:
(Even though I installed conda packages first)

1. sudo apt-get –purge remove nvidia-*
2. sudo apt-get autoremove
3. sudo add-apt-repository ppa:graphics-drivers/ppa
4. sudo apt-get update
5. sudo apt-get install nvidia-375
6. sudo reboot
7. wget https://repo.anaconda.com/archive/Anaconda3-5.2.0-Linux-x86_64.sh 7.1. bash Anaconda3-5.2.0-Linux-x86_64.sh
   While installation,when asked about adding path,say yes.
8. source ~/.bashrc
9. conda update conda
10. conda update anaconda
11. conda update python
12. conda update –all
13. conda create –name tf-gpu
14. source activate tf-gpu
15. conda install tensorflow-gpu
    I tried this and it worked. So now if you want to check if gpu device is recognized or not:
16. python
    16.1. import tensorflow as tf

    with tf.device('/gpu:0'):
    a = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[2, 3], name='a')
    b = tf.constant([1.0, 2.0, 3.0, 4.0, 5.0, 6.0], shape=[3, 2], name='b')  
    c = tf.matmul(a, b)  
    with tf.Session() as sess:  
    print (sess.run(c))
    

17. That’s it you are done with installation.
18. Reference