nouveau grabs the card instead of the NVIDIA driver (or a stale blacklist stops nvidia loading)
The core is right and the diagnostics are unusually good: lspci -k -d ::03xx is valid (pciutils lib/filter.c parse_hex_field accepts 'x' wildcards for the class field, and the Hyprland Multi-GPU wiki uses the identical form), /etc/modprobe.d overriding /usr/lib/modprobe.d is correct, and systemd-analyze cat-config modprobe.d is the right merged view. Three defects. (1) Direction B's headline symptom is misattributed: blacklist nvidia does NOT stop an explicit modprobe nvidia: blacklist only suppresses alias-driven autoload. could not insert 'nvidia': Operation not permitted (EPERM) comes from an install nvidia /bin/false line, /proc/sys/kernel/modules_disabled, or lockdown, not from a blacklist entry. A reader hunting for a bare blacklist line on that error will find nothing. (2) The quoted /usr/lib/modprobe.d/nvidia-utils.conf is stale: the current file (nvidia-utils 610.57.04) also carries options nvidia NVreg_UseKernelSuspendNotifiers=1 and options nvidia NVreg_TemporaryFilePath=/var/tmp. The blacklist/softdep half quoted is verbatim correct. (3) sudo env OMARCHY_ALLOW_DIRECT_PACMAN=1 pacman -S nvidia-utils is cargo-cult: I read the guard, and it only trips when both a sync and a sysupgrade flag are present, so a plain pacman -S never needed the escape hatch.
Adding the NVIDIA modules to MODULES= (early KMS) is documented by both Arch and the Hyprland wiki to break resume-from-hibernation. The machine cold-boots instead of restoring. If you hibernate, skip that step. Removing kms from HOOKS costs you the early framebuffer and a themed Plymouth/LUKS prompt. Always rebuild the initramfs *and* regenerate the boot entries before rebooting. A UKI built from a half-edited mkinitcpio.conf boots to a black screen, and on Omarchy 4 you then need the Limine snapshot entry to get back.
Symptom
Two mirror-image complaints.
Direction A is "I installed the NVIDIA driver but nothing uses it": lspci -k says Kernel driver in use: nouveau, nvidia-smi fails, and glxinfo -B reports a Mesa/NVK/llvmpipe renderer. Games run at single-digit FPS.
Direction B is the reverse: nvidia-smi fails with NVIDIA-SMI has failed because it couldn't communicate with the NVIDIA driver, and journalctl -b shows the module was refused:
modprobe: ERROR: could not insert 'nvidia': Operation not permitted
or sudo modprobe -v nvidia prints install /bin/false / reports the module is blacklisted, while /etc/modprobe.d/ looks empty to you.
Cause
nouveau and nvidia both claim the same PCI device. Whichever binds first wins. The blacklist that separates them ships with nvidia-utils at /usr/lib/modprobe.d/nvidia-utils.conf, which currently contains blacklist nouveau, blacklist nova_core, blacklist nova_drm, softdep nvidia post: nvidia-uvm nvidia-drm, plus two options nvidia NVreg_* lines.
Direction A (nouveau wins) happens when nvidia-utils is not installed (you installed only nvidia-dkms), or the initramfs was built before it was and the kms hook still pulls nouveau in during early boot, or the NVIDIA module failed to build for the running kernel.
Direction B (nvidia refused) is where the original cause overreached. A plain blacklist nvidia line only suppresses automatic, alias-driven loading. It does not block an explicit modprobe nvidia, and it never produces Operation not permitted. That EPERM refusal comes from one of: an install nvidia /bin/false line in /etc/modprobe.d/ (the form old "disable the dGPU" howtos actually use, and the only modprobe.d form that defeats explicit loading), /proc/sys/kernel/modules_disabled=1, or kernel lockdown/module-signature enforcement. A module_blacklist=/modprobe.blacklist= kernel parameter likewise only blocks autoload and is invisible to modprobe.d inspection. So on Direction B, check for install ... /bin/false and for enforcement, not just for a blacklist line, because /etc/modprobe.d/ genuinely can look empty of relevant blacklists while nvidia is still being refused.
Fix
**1. Find out who owns the card and what is actually blocking what.**
lspci -k -d ::03xx # "Kernel driver in use:" line is the answer
lsmod | grep -E 'nouveau|nvidia|nova'
# merged view of EVERY modprobe.d fragment, in priority order:
systemd-analyze cat-config modprobe.d | grep -nE 'nouveau|nvidia|nova'
grep -rnE 'blacklist (nouveau|nvidia|nova)|^install (nvidia|nouveau)' /etc/modprobe.d/ /usr/lib/modprobe.d/
# the command line overrides modprobe.d entirely - check it too:
grep -oE 'module_blacklist=[^ ]*|modprobe\.blacklist=[^ ]*|nomodeset' /proc/cmdline
sudo modprobe -v nvidia # shows exactly what modprobe would do
If modprobe nvidia fails with Operation not permitted, that is NOT a plain blacklist. Check these three instead:
cat /proc/sys/kernel/modules_disabled # want: 0
cat /sys/kernel/security/lockdown # want: [none] integrity confidentiality
cat /sys/module/module/parameters/sig_enforce # want: N
**Direction A: nouveau is winning.** Install the package that owns the blacklist, then rebuild the initramfs:
# Omarchy 4 (idiomatic; the update guard only blocks -Syu, so plain -S is fine either way):
omarchy-pkg-add nvidia-utils
# plain Arch/EndeavourOS/CachyOS:
sudo pacman -S nvidia-utils
# optional but decisive: load the NVIDIA modules early instead of nouveau
sudoedit /etc/mkinitcpio.conf
# MODULES=(nvidia nvidia_modeset nvidia_uvm nvidia_drm)
# (on an Intel iGPU + NVIDIA dGPU laptop put i915 FIRST: MODULES=(i915 nvidia ...))
sudo limine-mkinitcpio && sudo limine-update # Omarchy 4
# or, on plain mkinitcpio systems:
sudo mkinitcpio -P
sudo reboot
If you would rather not use early KMS, you can instead keep nouveau out of the image by removing kms from the HOOKS array in /etc/mkinitcpio.conf and rebuilding, the alternative the Arch NVIDIA page documents.
**Direction B: something is stopping nvidia.** Remove whatever the greps above named:
# a blacklist/install fragment (use the real filename you found):
sudo rm /etc/modprobe.d/blacklist-nvidia.conf
# if it was on the command line, on Omarchy 4 (note tee -a, so you do not
# clobber an existing zz-local.conf):
sudoedit /etc/limine-entry-tool.d/zz-local.conf # remove module_blacklist=nvidia...
sudo limine-mkinitcpio && sudo limine-update
sudo reboot
If instead modules_disabled was 1, or lockdown/sig_enforce was enforcing, this is not a blacklist problem at all. See the Secure Boot / module-signature record.
Verify
lspci -k -d ::03xx shows Kernel driver in use: nvidia. cat /sys/module/nvidia_drm/parameters/modeset returns Y. nvidia-smi prints the device table. lsmod | grep nouveau is empty.
Sources
- https://wiki.archlinux.org/title/NVIDIA
- https://wiki.archlinux.org/title/Kernel_module
- https://gitlab.archlinux.org/archlinux/packaging/packages/nvidia-utils/-/raw/main/nvidia-utils.conf
- https://github.com/hyprwm/hyprland-wiki/blob/main/content/Nvidia/_index.md
- https://github.com/basecamp/omarchy/blob/quattro/bin/omarchy-update-pacman-guard