Installation was done on: April 12th, 2018. Guide prepared on April 14th, 2018
Recently, I upgraded my NVidia Graphics Card, old drivers were not compatible with the new GPU. I had a tough time to setup Tensorflow-GPU on Ubuntu. Hence, decided to make a comprehensive installation instruction guide.
- Graphics Card: MSI Duke GTX 1080
- Ubuntu 16.04.4 LTS -- Download from here
- Ubuntu Kernel - 4.13.0-38-generic (You can check this using
uname -r) - GCC 5.4.0 20160609
- Cuda 9.0
- CudaNN 7.0
- Python 2.7.12
- Python 3.5.2
- Tensorflow 1.7.0
Make a bootable USB
On Windows
On Ubuntu
Note: If you are using windows to make a bootable USB, make sure that when you are formatting your USB Stick you select 4096 Bytes as the allocation size
Boot from the USB drive, follow the instructions and install Ubuntu on your machine.
Note: If your system uses UEFI instead of BIOS, you need to disable Secure boot. This is mandatory for you to use third party drivers.
Update all the already installed packages.
sudo apt-get update
sudo apt-get upgrade
Make sure python, python3 are already installed.
python --version
python3 --version
else Install those
sudo apt-get install python-dev python3-dev
Install the required packages.
sudo apt-get install openjdk-8-jdk git python-numpy python3-numpy build-essential python-pip python3-pip python-virtualenv swig python-wheel libcurl3-dev vim
The already installed drivers for your GPU wont work with Cuda, so one has to install the proprietary drivers. For doing that blacklist the existing Noveau drivers.
Create a file:
sudo vi /etc/modprobe.d/nouveau
Add the following lines in the above file:
blacklist nouveau
blacklist lbm-nouveau
options nouveau modeset=0
alias nouveau off
alias lbm-nouveau off
Update the initial RAM filesystem
echo options nouveau modeset=0 | sudo tee -a /etc/modprobe.d/nouveau-kms.conf
sudo update-initramfs -u
Reboot your machine:
sudo shutdown -r now
After reboot make sure the Nouveau drivers are blacklisted.
lsmod | grep nouveau
If nouveau drivers are still loaded troubleshoot the issue and then proceed with step6.
Purge NVidia drivers
sudo apt-get purge nvidia*
sudo reboot
Reboot your machine
sudo shutdown -r now
Install Cuda. Installing Cuda will install the required NVidia Drivers. (Version: Cuda 9.0.176-1)
curl -O http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/cuda-repo-ubuntu1604_9.0.176-1_amd64.deb
Download key to allow installation
sudo apt-key adv --fetch-keys http://developer.download.nvidia.com/compute/cuda/repos/ubuntu1604/x86_64/7fa2af80.pub
Extract the package and install
sudo dpkg -i ./cuda-repo-ubuntu1604_9.0.176-1_amd64.deb
sudo apt-get update
sudo apt-get install cuda-9-0
Restart your machine
sudo shutdown -r now
After restart make sure NVidia drivers are installed.
nvidia-smi
This should show the GPUs on your machine, driver version (390.30 in my case), etc. If this doesnt show up troubleshoot the issue and then proceed to step8.
Install CudNN. (Version: CudNN 7.0.5) Download the required version of CudaNN from here. You will have to create an account to download CudNN. Navigate to the path where the above file was downloaded and run the following.
sudo tar -xzvf cudnn-9.0-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*
Add the following lines to your ~/.bashrc
export PATH=/usr/local/cuda-9.0/bin${PATH:+:${PATH}}
export LD_LIBRARY_PATH=/usr/local/cuda-9.0/lib64${LD_LIBRARY_PATH:+:${LD_LIBRARY_PATH}}
Reload ~/.bashrc
source ~/.bashrc
Install tensorflow-gpu
- For python 2.7
pip install tensorflow-gpu
- For python 3.5
pip3 install tensorflow-gpu
Test tensorflow installation Open python in a terminal
python #for python 2.7
python3 #for python 3.5
Run the following commands.
import tensorflow as tf
sess = tf.Session()
If the installation was successful, you should see many lines with gpu and gpu model in them.