Arch Linux User Repository

@caiotbc: Thanks for the detailed set of information!

From what I could see:

• nvidia module for the kernel 6.4.8 was built (which means the kernel headers were indeed present at the time of compilation)

• at boot time, the system tries to load the nvidia kernel module, but there is some issue (udev-worker)

Aug 04 20:04:07 caio-arch-x470 (udev-worker)[255]: nvidia: Process '/usr/bin/bash -c 'if [ ! -c /dev/nvidiactl ]; then /usr/bin/mknod -Z -m 666 /dev/nvidiactl c $(grep nvidia-frontend /proc/devices | cut -d \  -f 1) 255; fi'' failed with exit code 1.

• X server loads the nvidia module, tries to configure it, but the nvidia module refuses to accept the video card:

[    12.541] (II) NVIDIA GLX Module  470.199.02  Thu May 11 11:49:03 UTC 2023
[    12.541] (II) NVIDIA: The X server supports PRIME Render Offload.
[    15.623] (EE) NVIDIA(GPU-0): The NVIDIA GPU at PCI:6:0:0 is not supported by the 470.199.02
[    15.623] (EE) NVIDIA(GPU-0):     NVIDIA driver.
[    15.623] (EE) NVIDIA(GPU-0): Failed to initialize the NVIDIA graphics device!
[    15.623] (EE) NVIDIA(0): Failing initialization of X screen

• nvidia device is indeed at the PCI location refused by the driver

06:00.0 VGA compatible controller [0300]: NVIDIA Corporation GK104 [GeForce GTX 680] [10de:1180] (rev a1) (prog-if 00 [VGA controller])
        Subsystem: eVga.com. Corp. GK104 [GeForce GTX 680] [3842:2682]
        Physical Slot: 0-6
        Flags: bus master, fast devsel, latency 0, IRQ 60
        Memory at 80000000 (32-bit, non-prefetchable) [size=16M]
        Memory at 706000000000 (64-bit, prefetchable) [size=128M]
        Memory at 706008000000 (64-bit, prefetchable) [size=32M]
        I/O ports at c000 [size=128]
        Expansion ROM at <ignored> [disabled]
        Capabilities: <access denied>
        Kernel driver in use: nouveau
        Kernel modules: nouveau

At this point, I would be curious to experiment with the package nvidia-prime, since I saw that Prime offloading is supported.

There is also optimus-manager, which is supposed to help for configurations like yours.

Also, could you share the contents of your xorg.conf file ? The trace from Xorg.0.log is surprisingly quiet about the parameters read from xorg.conf which makes me believe there is no such file and, perhaps, you might need one.

Since my config is really different from yours, I cannot experiment or copy elements from it, and we need to take it slowly to find the real reason why the driver rejects your card.

Edit: corrected some markdown and added reference to optimus-manager.

@SoftExpert Thanks for the quick response. I would like to point out that this GPU is usually attached to a VM using PCI passthrough, on my first comment when I say that "no other GPUs are installed" I mean that no other GPUs are attached to the same guest. I do however have an AMD GPU running on the host, which is also arch linux, same kernel version (6.4.8-arch1-1). I've run this setup for a long time (debian host/debian guest) and saw no difference between bare metal/virtualization so it's something I usually don't even mention, however just to make sure it wasn't the issue I did try to install the same driver on the bare metal arch install and found the same issue. I've compiled a text file with the informations you asked and some others that I found relevant, it got pretty big so I've uploaded it to pastebin and divided into sections that you can CTRL+F to using the titles below:
1 - neofetch output (running nouveau)
2 - makepkg of the package
3 - install via pacman
4 - contents of /var/lib/dkms/nvidia/
5 - checking kernel and headers by trying to reinstall them
6 - lspci (running nouveau before reboot)
7 - output of /var/log/Xorg.0.log
8 - output of journalctl -b0

https://pastebin.com/NCZQnLYm

Edit: learning markdown line breaks :)

@caiotbc: According to Nvidia your card is supported by this driver.

The only issue you must have is that the kernel modules could not be compiled, which can happen for several reasons.

Is it possible to list the contents of /var/lib/dkms/nvidia ? Also, can you tell us which version of the kernel you are using and if you installed the corresponding kernel headers ?

Edit: corrected link markdown.

Hi! I'm a recent Arch user and am having trouble getting this package to work. It downloads and installs fine, however after rebooting X refuses to start. My GPU is a GK104 GTX680 from EVGA, this is a desktop computer and no other GPUs are installed. Nouveau drivers work out of the box. The X server log shows the following error lines:

[ 15.623] (EE) NVIDIA(GPU-0): The NVIDIA GPU at PCI:6:0:0 is not supported by the 470.199.02 [ 15.623] (EE) NVIDIA(GPU-0): NVIDIA driver. [ 15.623] (EE) NVIDIA(GPU-0): Failed to initialize the NVIDIA graphics device!

I'm confused since this GPU was working with the 470 drivers on a previous debian install, and checking the links on arch wiki my model should be supported by this driver package. What can I do to fix this?

I updated the package with an additional patch that works with kernel 6.5-rc4. Usually these patches will work with the final release as well. The package will now compile with kernels 6.0.x up to (and including) 6.5.x.

running dkms with the correct driver version instead of a 3 year old one:

dkms install --no-depmod nvidia/470.199.02 -k 6.4.7-zen1-1-zen

works flawlessly :)

@Sinclaus: So first of all, one exclamation mark is enough :D

Secondly, you are compiling 470.63.01 - that driver is ancient, from like 2021, what does it have to do with this package? we are on 470.199.02 atm

Error on compile!!!!!!

Standard update process and compilation by

dkms install --no-depmod nvidia/470.63.01 -k 6.4.7-arch1-1

gives many errors like this: .... make[2]: *** [scripts/Makefile.build:252: /var/lib/dkms/nvidia/470.63.01/build/nvidia/nv-p2p.o] Error 1 In file included from /var/lib/dkms/nvidia/470.63.01/build/common/inc/nv-linux.h:15, from /var/lib/dkms/nvidia/470.63.01/build/common/inc/nv-pci.h:15, from /var/lib/dkms/nvidia/470.63.01/build/nvidia/nv-pci.c:13: /var/lib/dkms/nvidia/470.63.01/build/common/inc/nv.h:22:12: fatal error: stddef.h: No such file or directory 22 | #include <stddef.h> // NULL | ^~~~~~~~~~ compilation terminated. make[2]: *** [scripts/Makefile.build:252: /var/lib/dkms/nvidia/470.63.01/build/nvidia/nv-pci.o] Error 1 In file included from /var/lib/dkms/nvidia/470.63.01/build/nvidia/nv-i2c.c:15: /var/lib/dkms/nvidia/470.63.01/build/common/inc/os-interface.h:27:10: fatal error: stdarg.h: No such file or directory 27 | #include <stdarg.h> | ^~~~~~~~~~

@frdbr: what SoftExpert said, basically. We need way more info because the chance is high it is due to your specific config/machine rather than the driver, because it works for the others. I myself am running 4 FHD screens in parallel successfully with this driver version and a GTX780.

@frdbr: You do realize the given information has the same value as "It does not work"... There are a thousand different reasons for which your problem might happen: - electrical: loose connection, bad cables, etc - firmware: in some situations, bugs in the display firmware can work against a correct connection - software: bad config, bugs in the kernel, bugs in the driver, bugs in the display manager of your desktop environment - any combination of the previous reasons It is worth checking with a different kernel, turning off (unplug) the monitor, scrub the X11 config file (if you try to use Wayland with Nvidia - it's a feature, not a bug), inspect the output from xrandr, check the version of your desktop libraries, etc... And, if you like some help, share all that information with us ...