Install NVIDIA-driver (GeForce RTX 2070 SUPER), cuda 10.1, cudnn 7.6 on Ubuntu 18.04.3 LTS + python environment construction

Introduction

You will be setting up the GPU, and I will keep the procedure as a memorandum. In this article, we have installed ** NVIDIA-driver, cuda, cudnn ** and created a Python virtual environment using ** pyenv, virtualenv **. I hope it will be helpful as much as possible.

environment

• ubuntu: 18.04.3 LTS
• Graphics: GeForce RTX 2070 SUPER
• Ndivier-driver: 430.50
• cuda: 10.1
• cudnn: 7.6
• Python: 3.6.5

1. Before installing NVIDIA-driver

Install vim (because I personally want to use vim)

$ sudo apt update
$ sudo apt install vim

I want to use jj with vim, so edit ~ / .vimrc. (Temporarily I will only do this setting.)

~/.vimrc

set number
inoremap<silent> jj <ESC>

(1) Disabling Nouveau

First, disable Nouveau. When it comes to Nvidia graphics cards, a driver called Nouveau is set by default, so add Nouveau to the blacklist.

Check if Nouveau is used

$ lsmod | grep -i nouveau

Creating a blacklist

$ sudo vim /etc/modprobe.d/blacklist-nouveau.conf

/etc/modprobe.d/blacklist-nouveau.conf

blacklist nouveau
options nouveau modeset=0

Execute the following command and confirm that nouveau is disabled

OK if the display resolution is low

$ sudo update-initramfs -u
$ sudo reboot 

(2) Fixing the kernel

If you do not fix the kernel version of the nvidia driver, it seems that the dependency with the driver may be broken when upgrading. So, fix the kernel.

Install aptitude

$ sudo apt install aptitude

aptitude: A text-based interface to the Debian GNU / Linux package system that allows you to view a list of packages and perform package management tasks such as installing, updating, and removing packages.

Check kernel version

$ aptitude show linux-generic

Write the contents confirmed above to the following file (rewrite only the version)

$ cd /etc/apt/preferences.d
$ sudo vim linux-kernel.pref

.linux-kernel.pref

Package: linux-generic
Pin: version 4.15.0.64.66
Pin-Priority: 1001

Package: linux-headers-generic
Pin: version 4.15.0.64.66
Pin-Priority: 1001

Package: linux-image-generic
Pin: version 4.15.0.64.66
Pin-Priority: 1001

That's all for fixing the kernel.

2. Install NVIDIA driver

Check if nouveau is disabled

$ lsmod | grep -i nouveau

If nothing is displayed, it can be disabled

Install all the tools required for development such as gcc and make

$ sudo apt install build-essential

Add repository

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

Shows installable devices

$ ubuntu-drivers devices

Select the driver you want to install and install

$ sudo apt install nvidia-driver-430
$ sudo reboot

Check if the driver is installed

$ nvidia-smi

OK if the usage status of GPU etc. is displayed

Another way to install NVIDIA-driver

I wrote an article before, so please refer to that. I installed the NVIDIA driver on Ubuntu 17.10.

3. Install cuda

Please download cuda from here (https://developer.nvidia.com/cuda-toolkit-archive). Be sure to check the version of cuda and driver.

cuda installation

$ sudo apt update
$ sudo dpkg -i cuda-repo-ubuntu1804-10-1-local-10.1.105-418.39_1.0-1_amd64.deb
$ sudo apt-key add /var/cuda-repo-10-1-local-10.1.105-418.39/7fa2af80.pub 
$ sudo apt update
$ sudo apt install cuda-10-1

Edit ~ / .bashrc (setting environment variables)

~/.bashrc

export PATH="/usr/local/cuda-10.1/bin:$PATH"
export LD_LIBRARY_PATH="/usr/local/cuda-10.1/lib64:$LD_LIBRARY_PATH"

Pass through PATH

$ source ~/.bashrc

Version confirmation

$ nvcc -V
$ nvidia-smi

4. Install cudnn

Please download the cudnn installation file from here (https://developer.nvidia.com/rdp/cudnn-download). You will need to register for an account.

installation of cudnn

$ tar xvf cudnn-10.1-linux-x64-v7.6.5.32.tgz
$ sudo cp -a cuda/include/cudnn.h /usr/local/cuda/include/
$ sudo cp -a cuda/lib64/libcudnn* /usr/local/cuda/lib64/
$ sudo chmod a+r /usr/local/cuda/include/cudnn.h /usr/local/cuda/lib64/libcudnn* 
$ sudo reboot

Test if cudnn was installed

$ sudo dpkg -i libcudnn7-doc_7.6.5.32-1+cuda10.1_amd64.deb
$ cp -r /usr/src/cudnn_samples_v7/ $HOME
$ cd $HOME/cudnn_samples_v7/mnistCUDNN
$ make clean && make
$ ./mnistCUDNN

OK if Test passed! Is displayed

cudnn version check command

$ cat /usr/local/cuda/include/cudnn.h | grep CUDNN_MAJOR -A 2

5. Python environment construction

I got a dependency error so I repaired it

$ sudo apt --fix-broken install

Install required packages for Python

$ sudo apt install git gcc make openssl libssl-dev libbz2-dev libreadline-dev libsqlite3-dev zlib1g-dev

(1) Installation of pyenv

** pyenv **: A command line tool for Python version control that makes it easy to install and switch between multiple versions of Python.

$ git clone https://github.com/pyenv/pyenv.git ~/.pyenv
$ echo 'export PYENV_ROOT="$HOME/.pyenv"' >> ~/.bashrc 
$ echo 'export PATH="$PYENV_ROOT/bin:$PATH"' >> ~/.bashrc 
$ echo 'eval "$(pyenv init -)"' >> ~/.bashrc
$ source ~/.bashrc

List of installable packages

$ pyenv install -l

Python installation

$ pyenv install 3.6.5

Check the installed version

$ pyenv versions

Version switching

$ pyenv global 3.6.5

(2) virtualenv

** virtualenv **: Software that can create a separate Python virtual environment in one system. You can easily create a virtual environment with commands.

Install virtualenv

$ pip install virtualenv

Version confirmation

$ virtualenv --version

Create virtual environment

$ virtualenv (PJ name) --no-site-packages

Enable virtual environment

$ . (PJ name)/bin/activate

Disable virtual environment

$ deactivate

At the end

I installed from Ubuntu this time, but the installation screen did not appear at first. The following settings were required on the BIOS screen.

--Secure Boot: os type UEFI mode ⇒ non-UEFI mode --CSM: Start CSM ⇒ Enabled

It took a while, but it was a good experience to set up this time because I gained some knowledge.