Skip to content

Latest commit

 

History

History
168 lines (100 loc) · 3.94 KB

nvidia.md

File metadata and controls

168 lines (100 loc) · 3.94 KB
Raw

Guide for complete nvidia and tensorflow setup on arch linux using System76 Drivers

~ Nageen

Source

Setting up nvidia drivers and making tensorflow to work on gpu on linux system is always mind boggling task. I have many times reinstalled/formatted my laptop just to get it working. No matter what there is always something that messes up while setting up working environment for tensorflow on grphic card.

In this post, I will explain an easy to perform procedure for installing everything on Arch Linux so that you can finally execute tensorflow programs on GPU.

Note: I am assuming that you have working Arch Linux installed with any Desktop Environment(Gnome, KDE, i3 etc) with no extra packages related to nvidia or tensorflow(e.g Bumblebee, optimus, nvidia etc). Make sure you have yay installed. We will install system76-power from Arch User Repository(AUR). [ System76 have their own linux distribution Pop OS based on Ubuntu which have inbuilt nvidia support.]

So lets get started:


    yay -S system76-power system76-dkms
    sudo systemctl enable system76-power.service

This will take some time to install.

System76 Drivers require “ec_sys.write_support=1” argument to be passed for kernel to work. Open grub file with:


    sudo nano /etc/default/grub

And search for line starting with GRUB_CMDLINE_LINUX_DEFAULT and append this in it. It should look something like this:


    ...
    GRUB_CMDLINE_LINUX_DEFAULT="quiet splash ec_sys.write_support=1"
    ...

Then we will regenerate grub confab with this new setting using:


    sudo grub-mkconfig -o /boot/grub/grub.cfg

Install bumblebee and optimus:


    sudo pacman -S bumblebee primus
    sudo gpasswd -a $USER bumblebee
    sudo systemctl enable bumblebeed.service
    sudo pacman -S nvidia-dkms nvidia-settings

Reboot your system.

system76-power is used to switch the graphic cards between Intel and nvidia, but it need a reboot after the change.

Switch the graphics to Intel:


    sudo system76-power graphics intel

And to switch to nvidia, use:


    sudo system76-power graphics nvidia

After reboot, turn on graphic card and load nvidia drivers using:


    sudo system76-power graphics power on
    sudo modprobe nvidia-drm nvidia-modeset nvidia

To confirm if nvidia drivers have successfully loaded and working, use:


    primusrun glxinfo | grep -i renderer

Output should be something like :


    OpenGL renderer string: GeForce 940MX/PCIe/SSE2

And to turn off:


    sudo rmmod nvidia-drm nvidia-modeset nvidia
    sudo system76-power graphics power off

For easy use you can you bash alias as follow:


    alias nvidia-settings="optirun -b none nvidia-settings -c :8 "
    alias nvidia-on="sudo system76-power graphics power on ; sudo modprobe nvidia-drm nvidia-modeset nvidia"
    alias nvidia-off="sudo rmmod nvidia-drm nvidia-modeset nvidia ; sudo system76-power graphics power off"

Now you can install tensorflow from directly arch linux official repository


    sudo pacman -S python-tensorflow-cuda

If everything goes fine. You are ready to run tensorflow programs with GPU now.

If you want to install nvidia drivers and tensorflow from from official nvidia website follow this post.

If you have any query, Comment below.

References:

  1. https://wiki.archlinux.org/index.php/NVIDIA
  2. https://ebobby.org/2018/07/15/archlinux-on-oryp4/
  3. https://www.archlinux.org/packages/community/x86_64/python-tensorflow-cuda/