Hyprland on NVIDIA (Arch Linux, 2026): A Setup That Actually Works

Most "Hyprland + NVIDIA" guides online were written in 2023 and are full of steps you no longer need — and a couple that now do harm. This is the setup I run today on an RTX 4050 laptop with Hyprland 0.56 and the nvidia-open driver, with every step checked against what current Arch packages actually do.

If you haven't picked a compositor yet, start with my Hyprland vs Sway vs i3 comparison.

Step 1: pick the right driver package

Arch's driver packaging changed in December 2025. NVIDIA's 590 driver dropped support for Pascal (GTX 10-series) and older cards, and Arch's main packages switched to NVIDIA's open kernel modules (nvidia became nvidia-open, nvidia-dkms became nvidia-open-dkms). Find your GPU with lspci | grep -Ei "vga|3d", then:

RTX 50-series cards only work with the open modules. Use the -dkms variant if you run more than one kernel — for example the linux-lts fallback I recommend in keeping Arch stable.

# Turing and newer, stock kernel
sudo pacman -S nvidia-open nvidia-utils lib32-nvidia-utils libva-nvidia-driver

# or, with several kernels
sudo pacman -S nvidia-open-dkms linux-headers linux-lts-headers \
  nvidia-utils lib32-nvidia-utils libva-nvidia-driver

lib32-nvidia-utils needs the multilib repo (for Steam and Wine). libva-nvidia-driver gives you hardware video decoding in browsers and mpv.

GTX 1060 / 1070 / 1080 owners: if your desktop went black after an update around Christmas 2025, this is why. Boot to a TTY (Ctrl+Alt+F2), remove nvidia / nvidia-dkms, install nvidia-580xx-dkms from the AUR with your AUR helper, and reboot.

Step 2: check what's already done for you

Old guides tell you to add nvidia_drm.modeset=1 and nvidia_drm.fbdev=1 to your kernel command line. Current drivers turn both on by default. Check rather than guess:

cat /sys/module/nvidia_drm/parameters/modeset   # should print Y
cat /sys/module/nvidia_drm/parameters/fbdev     # should print Y

If either prints N, add the option in /etc/modprobe.d/nvidia.conf:

options nvidia_drm modeset=1 fbdev=1

Arch's nvidia-utils also ships /usr/lib/modprobe.d/nvidia-utils.conf (blacklists nouveau, enables kernel suspend notifiers), so you don't need to write any of that yourself.

Early KMS (adding nvidia nvidia_modeset nvidia_uvm nvidia_drm to MODULES=() in /etc/mkinitcpio.conf) is optional. Do it only on a desktop where the NVIDIA card drives the screen and you see a black screen or a wrong resolution at boot. On a hybrid laptop the Intel/AMD GPU drives the internal display, so leave it out.

Step 3: environment variables

In Hyprland 0.55+ the config is Lua (~/.config/hypr/hyprland.lua). If you're still on hyprland.conf, read my Lua migration guide first.

Desktop, NVIDIA drives the display:

hl.env("LIBVA_DRIVER_NAME", "nvidia")
hl.env("__GLX_VENDOR_LIBRARY_NAME", "nvidia")
hl.env("NVD_BACKEND", "direct")
hl.env("ELECTRON_OZONE_PLATFORM_HINT", "auto")

Hybrid laptop (Intel/AMD + NVIDIA): set only ELECTRON_OZONE_PLATFORM_HINT. Setting the first two globally pushes every app onto the NVIDIA GPU, which drains the battery and breaks some apps. Run individual apps on the dGPU with prime-run instead (below).

Things you should not set any more:

• WLR_NO_HARDWARE_CURSORS — deprecated; hardware cursors work
• GBM_BACKEND=nvidia-drm — not needed with current drivers
• XDG_SESSION_TYPE=wayland — your session launcher sets it

Check video decoding with vainfo (from libva-utils). It should list H.264/HEVC/AV1 profiles under the NVIDIA driver.

Step 4: suspend and resume

This used to be the worst part of NVIDIA on Linux. With nvidia-open and Arch's defaults it mostly just works: nvidia-utils sets NVreg_UseKernelSuspendNotifiers=1, and with the open modules that handles saving video memory across suspend automatically.

On the legacy nvidia-580xx driver you still need the old setup:

# /etc/modprobe.d/nvidia-power.conf
options nvidia NVreg_PreserveVideoMemoryAllocations=1 NVreg_TemporaryFilePath=/var/tmp

sudo systemctl enable nvidia-suspend.service nvidia-hibernate.service nvidia-resume.service

If you wake up to corrupted windows or a frozen screen, that's the first place to look: journalctl -b -1 | grep -i nvidia shows what happened in the previous boot. (More on reading logs in my systemd and journalctl guide.)

Step 5: hybrid laptops

Most gaming laptops have two GPUs. The integrated one runs the desktop and the NVIDIA one wakes up on demand.

Run an app on the NVIDIA GPU:

sudo pacman -S nvidia-prime
prime-run steam
prime-run blender

nvidia-smi lists which processes are using the dGPU right now. Useful when the fans won't stop.

Switching modes. EnvyControl (AUR: envycontrol) switches between integrated (dGPU off, best battery), hybrid (default) and nvidia (dGPU drives everything). I keep mine on hybrid and use prime-run for games.

External monitors. On many laptops the HDMI port is wired directly to the NVIDIA GPU. If an external screen feels laggy in hybrid mode, every frame is being copied between GPUs. Either switch to nvidia mode while docked, or tell Hyprland which GPU to prefer with AQ_DRM_DEVICES. Card numbers (/dev/dri/card0, card1) can swap between boots, and the stable /dev/dri/by-path/ names contain colons, which this variable uses as a separator. So create stable symlinks with a udev rule (the PCI addresses come from lspci):

# /etc/udev/rules.d/99-gpu-names.rules
KERNEL=="card*", KERNELS=="0000:00:02.0", SUBSYSTEM=="drm", SUBSYSTEMS=="pci", SYMLINK+="dri/igpu"
KERNEL=="card*", KERNELS=="0000:01:00.0", SUBSYSTEM=="drm", SUBSYSTEMS=="pci", SYMLINK+="dri/dgpu"

-- first entry is the primary renderer
hl.env("AQ_DRM_DEVICES", "/dev/dri/dgpu:/dev/dri/igpu")

Troubleshooting cheat sheet

The short version

1. Install the package that matches your GPU generation.
2. Verify modeset/fbdev are Y instead of copy-pasting kernel parameters.
3. Desktop: four env vars. Laptop: one, and prime-run for the rest.
4. Suspend is handled on nvidia-open; the old services are only for 580xx.

NVIDIA on Wayland went from "don't" to "boring" over the last two years, and boring is what you want from a GPU driver.