NVIDIA on Omarchy

Omarchy has real NVIDIA support, not an afterthought. The installer detects your card, picks a driver branch by GPU generation, and configures early kernel mode setting. On a desktop with a single NVIDIA card, the open reports are about what happens at update time and in packaging, not about the driver setup itself. On a hybrid laptop, where an Intel or AMD iGPU drives the panel and the NVIDIA chip sits behind it, most of the open complaints live.

This page was checked against v4.0.4, released 2026-09-15, using the v4.0.4 source tree. The issue counter in the sidebar covers every NVIDIA-matching issue and discussion in the tracker, open and closed, so treat it as a measure of traffic rather than of current breakage.

Status on 4.0.4

Which driver you get depends on the PCI device ID of your card.

• Turing or newer, meaning device ID 0x1e00 and above, roughly GTX 16 series and RTX 20 series onward. You get nvidia-open-dkms, nvidia-utils, lib32-nvidia-utils and libva-nvidia-driver. This is the supported path.
• Maxwell, Pascal and Volta, device IDs 0x1340 through 0x1dff, roughly GTX 700 later parts through GTX 10 series and Titan V. You get the legacy nvidia-580xx-dkms branch. Omarchy’s own repos carry 580.178.04 on the stable, rc and edge channels.
• Kepler and older. The installer prints that there is no compatible driver and points at the Arch wiki. You stay on nouveau, and the software cursor fix below is the only NVIDIA-specific thing Omarchy does for you.

Vulkan comes from nvidia-utils on NVIDIA. The separate vulkan.sh installer only handles Intel, AMD and Apple.

What Omarchy does automatically

Detects from sysfs, not lspci. omarchy-hw-nvidia, omarchy-hw-nvidia-gsp and omarchy-hw-nvidia-without-gsp read cached IDs under /sys/bus/pci/devices. The 4.0.0 notes explain why: an lspci call touches PCI config space, which pulls a runtime-suspended dGPU out of D3cold, and on a hybrid laptop that wake by itself blows through the 1.5 seconds Hyprland allows for loading its config.

Configures early KMS. It writes options nvidia_drm modeset=1 to /etc/modprobe.d/nvidia.conf and MODULES+=(nvidia nvidia_modeset nvidia_uvm nvidia_drm) to /etc/mkinitcpio.conf.d/nvidia.conf.

Drops the kms hook where it is dead weight. Since 4.0.0, omarchy_hooks.conf removes kms from HOOKS when nvidia_drm is early-loaded and NVIDIA owns every display controller. That cuts nouveau and about 100 MB of its GSP firmware out of the initramfs, the problem matjam measured in issue #6790 on a 256 MB unified kernel image. Hybrid machines keep kms, because the iGPU still needs it for the LUKS prompt. A migration rebuilds existing images once.

Sets session environment in Lua. default/hypr/nvidia.lua sets NVD_BACKEND=direct, LIBVA_DRIVER_NAME=nvidia and __GLX_VENDOR_LIBRARY_NAME=nvidia on GSP cards, and NVD_BACKEND=egl plus the GLX variable on older ones. In 3.x these were .conf env lines; in 4.x they are Lua, which matters if you are copying an old guide. See the conf to Lua migration.

Offers a hybrid GPU toggle. If omarchy-hw-hybrid-gpu reports hybrid, the Omarchy menu shows Hardware, Hybrid GPU, which runs omarchy-toggle-hybrid-gpu. That installs supergfxctl, writes /etc/supergfxd.conf, enables the daemon, and reboots. Switching to Integrated also installs a supergfxd startup delay, because starting it in Integrated mode races the display manager and can freeze the boot, and a force-igpu sleep hook that flips the card through Vfio back to Integrated after resume, and to Vfio before hibernate so the driver is never asked to freeze a powered-off dGPU.

Fixes the nouveau cursor. If you end up on nouveau, an installer appends no_hardware_cursors = true to your looknfeel.lua, because on many older cards nouveau never shows the hardware cursor plane, so under Hyprland you get no visible pointer at all.

Known problems

Issue	Models	Status	Fixed in
#12187 hard freeze about 5 seconds after login on the new default kernel	Lenovo hybrid, Intel iGPU plus RTX 4070 Max-Q, on prebuilt nvidia-open rather than DKMS	Open	not fixed
#5706 DKMS fails to build for a new kernel, SDDM login loop with no visible error	RTX 3090 Ti, 3060 Ti, 4060 Ti	Open, assigned, PR #6840 proposed	not fixed
#8215 NVDEC VA-API routing corrupts or kills browser video	Hybrid Intel plus RTX 3050, Meteor Lake plus RTX 4050, AMD Cezanne plus RTX 3050 Ti	Open, PR #7851 proposed	not fixed
#4901 cross-GPU DMA-BUF import fails, eglCreateImage failed with 0x00003009	Hybrid Intel plus RTX 3050, AMD plus NVIDIA confirmations	Open, PR #5312 proposed	not fixed
#12129 black screen on resume from suspend	ASUS TUF F17 FX707VI, RTX 4070 Mobile, kernel 7.2.5-3-omarchy	Open, no comments yet	not fixed
#12054 Sunshine NVENC probe fails, falls back to software x264	RTX 2080 Ti, single report	Open, no comments yet	not fixed
#6790 about 100 MB of unused nouveau GSP firmware in every initramfs	All NVIDIA-only installs	Closed	4.0.0
#1776 dGPU never reaches D3cold, battery drains	Hybrid Intel plus NVIDIA laptops	Closed 2026-02-01 as completed, thread is a manual setup guide	no linked fix

Two of these deserve detail.

The kernel module gap. 4.0.4 ships the bespoke linux-omarchy kernel to everyone and makes it the default boot entry. In issue #12187 the reporter was on the prebuilt nvidia-open package rather than the nvidia-open-dkms that Omarchy installs, so modules existed only under the stock kernel’s tree. Every boot on 7.2.5-3-omarchy froze about five seconds after login, with systemd-modules-load failing to find nvidia_uvm and supergfxd reporting nvidia_drm missing. The stock linux entry worked with identical userspace. Issue #5706 is the same shape from a different angle: DKMS failed to build, mkinitcpio printed module not found: 'nvidia', and the result was a login loop that gives the user no error at all.

The hybrid video decode regression. Omarchy 4.0.0 had a broken o.shell_succeeds helper, so the NVIDIA branch of nvidia.lua never ran. 4.0.1 fixed the helper, which switched LIBVA_DRIVER_NAME=nvidia on for the first time. On hybrid machines the iGPU composites while decode is forced to NVDEC, and the cross-GPU buffer import goes wrong. Reporters in issue #8215 describe stale frames rotating on a three frame cycle, wallpaper punch-through, and on one machine vaEndPicture failed, VA error: internal decoding error on every decoder instance. It reproduces with AMD iGPUs as well as Intel, so it is not Mesa iris specific.

Issue #2635, monitors showing no signal after wake, is often filed as an NVIDIA bug and the original report is a 3070. Read the thread before assuming: several confirmations are AMD Beelink mini PCs. Treat it as a display wake problem, not an NVIDIA one.

Fixes that work

Work in this order.

1. Confirm your driver branch. Run omarchy-hw-nvidia-gsp; echo $?. Zero means you should be on nvidia-open-dkms. Then check what is actually installed with pacman -Qs nvidia.
2. Make sure the package is the DKMS one. A prebuilt nvidia-open only has modules for the kernel it was built against. That is the trap in issue #12187.
3. Check modules exist for the kernel you boot. uname -r, then dkms status, then ls /usr/lib/modules/$(uname -r)/. If NVIDIA modules are missing there, do not reboot yet.
4. If you are already frozen or looping, pick the stock linux entry in Limine instead of the linux-omarchy one, or roll back with Snapper and Limine. Fix DKMS from a working session. See login loop or password not accepted and black screen after login.
5. For hybrid video corruption, add hl.env("LIBVA_DRIVER_NAME", "iHD") to ~/.config/hypr/hyprland.lua, using radeonsi instead on an AMD iGPU. Also run systemctl --user set-environment LIBVA_DRIVER_NAME=iHD so you do not have to log out, then restart the browser. Reporters note this variable affects video decode only, not Vulkan, OpenGL, NVENC or prime-run. One reporter in issue #4901 found that on a machine where Chromium renders through NVIDIA EGL, changing only the VA-API driver killed hardware decode entirely; the EGL vendor override in that thread has to go with it. More in Chromium flicker and hardware acceleration.
6. For battery drain on a hybrid laptop, try the Hybrid GPU toggle in the Omarchy menu and switch to Integrated. Details on the hybrid GPU page.
7. For resume failures, start with suspend will not resume and the suspend and sleep page, and say in your report whether the stock kernel behaves differently.

Report it

Run omarchy debug. It writes /tmp/omarchy-debug.log containing inxi -Farz, dmesg, this boot’s warnings and errors from the journal, and your full package list. Use --no-sudo to skip dmesg and --print to print instead of saving. See the command reference.

For an NVIDIA report, add four things the log does not make obvious: the output of lspci -k | grep -A3 -i vga, dkms status, the exact Limine entry you booted, and whether the problem also happens on the stock linux entry. That last one separates a kernel module gap from a driver bug and is the single most useful line you can include.

If your machine works, or works with one tweak, submit it so the model pages get more accurate.

Related

• NVIDIA drivers on Omarchy 4
• Hybrid GPU laptop black screen and AQ_DRM_DEVICES
• Kernel panic after update with Limine
• Steam or Proton game crashes
• Cursor invisible or wrong size
• Release channels and v4.0.4
• Omarchy manual: Troubleshooting and Gaming