4

I give my installation method below. Basically cannibalized from older posts on other drivers for older Ubuntu releases.

The purpose of this question is three fold:

  1. Learn the preferred way of installing the driver mentioned in the header of this post for the hardware in the header of this post.
  2. Learn pros/cons of my current approach
  3. Document my approach as there doesn't seem to be a clear post for my particular hardware/Ubuntu combination to aid others in a similar situation

Some context

I'm setting up a Deep Learning environment on my machine so am currently going through the pains of setting up the host machines driver and then will move on to installing nvidia-docker and so on.

My installation method:

  1. Download latest long lived release (367.44): http://www.nvidia.com/Download/driverResults.aspx/106780/en-us sudo apt install nvidia-modprobe
  2. Restart
  3. At login prompt: ctrl-alt-F1 for tty
  4. Login with credentials
  5. sudo service lightdm stop
  6. sudo init 3
  7. cd ~/Downloads
  8. chmod u+x [Nvidia installer here].run
  9. sudo init 3
  10. sudo ./[Nvidia installer here].run

    Ignore errors about pre-install script failing and 32 lib target install location not found.

    If there are errors saying that nouveau is still in use (after being blacklisted) then might be necessary to update the initramfs disk, so then do:

     sudo update-initramfs -u
    
     reboot
    

    A black screen will be present on reboot instead of normal Ubuntu splash screen.

    Repeat steps 3 and on. Ignore step 9 (only necessary once).

    Not sure if step 7 is needed again I had to do two iterations starting from step 3 and on since sudo init 3 seemed to actually start the X server so I was getting an “X server is running” error during install of .run file. So on second iteration (and consequently second reboot) I left out step 7 and it seemed to work.

  11. sudo service lightdm start
  12. sudo reboot
  13. Log in and check: System Settings -> Details -> Graphics Should read something like “GeForce ...” (assuming you have a GeForce series GPU) In my case: “GeForce GTX 970/PCIe/SSE2” NVIDIA X Server Settings -> NVIDIA Driver Version Should read “367.44”

Addendum:

I did see an Xorg error on startup/login: “Xorg crashed with SIGABRT in RRSetChanged()”. It seems that this is a known bug for Nvidia gpus and doesn’t seem to be critical.

Questions:

  1. I see that it "may" be possible to accomplish this installation via ppa:graphics-drivers
    Reference: NVidia GeForce GTX970 problem ubuntu 16.04

    Would this have accomplished the very same thing as my installation method and how does it compare (pros/cons) to mine above?

  2. Even though my installation above seems to have been successful, I'd like to know what pros/cons exist due to my approach and what (if any) issues might arise from my approach as I continue to install my Deep Learning environment (e.g. nvidia-docker, theano, keras, etc)?
John Cast
  • 141
  • The recommended way to update video driver is by selecting one from 'Additional Drivers'. Nvidia recommends using drivers by your Linux distribution. "Note that many Linux distributions provide their own packages of the NVIDIA Linux Graphics Driver... This may interact better with the rest of your distribution's framework, and you may want to use this rather than NVIDIA's..." (NVIDIA Linux Driver - Additional Information). Please keep this in mind, if you prefer the safer route. – David your friend Sep 02 '16 at 03:33
  • I would always prefer the graphics-drivers ppa over a driver download. Reasons and how-to here: http://askubuntu.com/a/61433/3940 – Takkat Sep 02 '16 at 06:32

0 Answers0