Arch Linux

jjcasmar

Member

Registered: 2019-11-24

Posts: 8

[Solved] Install nvidia driver for cuda, without opengl libs

I have a system with an AMD GPU which I use for rendering the desktop. I also have a nvidia card which I want to use for developing stuff in CUDA. If I try to install cuda through pacman, it install opengl libraries and override my desktop become unusable, as my monitors are connected to the AMD gpu. I have tried to install nvidia driver using the nvidia installer with the --no-opengl-libs option, but it fails to build the kernel module. Is there anyway I can install the nvidia driver from the package manager without installing opengl libraries?

Last edited by jjcasmar (2019-11-25 22:13:16)

seth

Member

From: Won't reply 2 private help req

Registered: 2012-09-03

Posts: 78,171

Re: [Solved] Install nvidia driver for cuda, without opengl libs

it install opengl libraries and override

That's not a thing.
OpenGL is handled by libglvnd since 2016, there's no symlinking / overriding going on - your issues lie elsewhere (or you're not in fact using archlinux)

Please post a complete system journal (w/ nvidia) and elaborate on "my desktop" (session type, display server)

---

How to upload text · How to boot w/o GUI · Disable Windows Fast-Start! · Fix your xinitrc

jjcasmar

Member

Registered: 2019-11-24

Posts: 8

Re: [Solved] Install nvidia driver for cuda, without opengl libs

With my desktop being useless I mean that it doesnt even shows the login window. I can just go to a terminal and remove nvidia driver to get a login screen again (after reboot). I am using plasma, sddm with X. I am pretty sure I am using arch linux :-)

How can I post the system journal?
Thanks

seth

Member

From: Won't reply 2 private help req

Registered: 2012-09-03

Posts: 78,171

Re: [Solved] Install nvidia driver for cuda, without opengl libs

sudo journalctl -b | curl -F 'f:1=<-' ix.io

Also post your xorg log (/var/log/Xorg.0.log) - I assume the nvidia GPU gets preferred and used, so try to ignore it, eg. by adding

Option "Ignore" "true"

to the nvidia device in /usr/share/X11/xorg.conf.d/10-nvidia-drm-outputclass.conf (nb. that this file likely gets overridden by future updates, but if it works, you can simply copy it to /etc/X11/xorg.conf.d)

---

How to upload text · How to boot w/o GUI · Disable Windows Fast-Start! · Fix your xinitrc

jjcasmar

Member

Registered: 2019-11-24

Posts: 8

Re: [Solved] Install nvidia driver for cuda, without opengl libs

The journalctl | curl command fails with Could  not resolve ix.io

The file you mentioned has an option PrimaryDevice yes. Should I remove it?

jjcasmar

Member

Registered: 2019-11-24

Posts: 8

Re: [Solved] Install nvidia driver for cuda, without opengl libs

Yep, removing the PrimaryDevice yes and adding the Ignore true options, works fine. xorg starts with the AMD GPU and I have a working desktop session again. Thanks!

seth

Member

From: Won't reply 2 private help req

Registered: 2012-09-03

Posts: 78,171

Re: [Solved] Install nvidia driver for cuda, without opengl libs

Great. Sorry for the delay.

Please always remember to mark resolved threads by editing your initial posts subject - so others will know that there's no task left, but maybe a solution to find.
Thanks.

---

How to upload text · How to boot w/o GUI · Disable Windows Fast-Start! · Fix your xinitrc