Hyprland will not start: 'aquamarine could not find a GPU', no seat, no DRM device
The three-way split of an ambiguous error is the right structure, and the seat half is solid: Arch Hyprland wiki line 19 says verbatim "Make sure to install the Polkit package, or start and enable seatd.service. As the lack thereof will cause Hyprland to fail to start", the seatd PKGBUILD comment confirms the seat group ('Allow users in the "seat" group to access seatd'), hyprpolkitagent is real (extra 0.1.3), the log path and lspci -k -d ::03xx are both correct, and 'never sudo Hyprland' is right. But the AQ_DRM_DEVICES advice is backwards on both counts. I read aquamarine src/backend/drm/DRM.cpp:194-236: it splits AQ_DRM_DEVICES on ':' and then calls std::filesystem::canonical() on *both* the values you gave it *and* each enumerated device path before comparing. Canonicalisation is exactly what makes a /dev/dri/by-path/... symlink work. It is not 'a known way to make it see nothing'. And the Multi-GPU wiki says the opposite of the record's fix: "Do not use the card1 symlink indicated here. It is dynamically assigned at boot and is subject to frequent change, making it unsuitable as a marker for GPU selection", i.e. prefer the stable by-path name over /dev/dri/cardN. Following the record's step 3 replaces a stable path with an unstable one. Second defect, Omarchy-specific: that same wiki page ends with "uwsm users are advised to export the AQ_DRM_DEVICES variable inside ~/.config/uwsm/env-hyprland, instead". Omarchy starts Hyprland under uwsm, so hl.env("AQ_DRM_DEVICES", ...) in hyprland.lua is the wrong place there. Third, minor: pacman -S drm_info fails. The package is drm-info.
Adding your user to input grants read access to every input device on the machine. That is a keylogging surface. Only do it if you actually need the seatd backend (logind users do not). Do not remove polkit to "test" the seatd path: polkit is what lets your session authenticate for mounts, network changes and reboots. Do not run sudo Hyprland as a workaround. It creates root-owned files in your $XDG_RUNTIME_DIR and in ~/.cache, which then break the normal session in ways that are tedious to unpick.
Symptom
Hyprland exits immediately back to the TTY. The log ends with:
[CRITICAL] m_pAqBackend was null! This usually means aquamarine could not find a GPU or encountered some issues. Make sure you're running either on a tty or on a Wayland session, NOT an X11 one.
and further up one or more of:
[ERR] libseat: failed to open a seat
[ERR] Failed to open a session
[ERR] libseat: Couldn't open device at /dev/dri/card1
[ERR] drm: No gpus in scanGPUs.
[ERR] drm: Found no gpus to use, cannot continue
[CRITICAL] No backend could be opened. Make sure there was a correct backend passed to CBackend, and that your environment supports at least one of them.
Older builds and other wlroots compositors phrase the same thing as Couldn't open a DRM device / Found 0 GPUs, cannot create backend.
Cause
Three separate failures produce nearly the same message, and the log line above tells you which:
- **No seat.** Hyprland takes DRM master through libseat, which needs either a valid systemd-logind session or a running seatd.
libseat: failed to open a seatmeans neither was available: launching over SSH, from inside tmux, withsudo Hyprland, from a display manager that never created a proper session, or on a system with no polkit package and seatd.service not enabled. The Arch Hyprland page is explicit: install a polkit package *or* enable seatd.service, otherwise Hyprland fails to start. - **No GPU visible.**
drm: No gpus in scanGPUs/Found no gpus to usemeans udev enumerated nothing with KMS. Causes:nomodeseton the kernel command line, the DRM driver blacklisted or not built for this hardware (see the i915/xe force_probe case), a VM with 3D acceleration disabled, or anAQ_DRM_DEVICESvalue naming a device node that does not exist or is not in the enumerated set. Note that a *symlink* is fine: aquamarine canonicalises both the paths you give it and each enumerated device path before comparing them, so /dev/dri/by-path/... resolves correctly, and the Hyprland Multi-GPU wiki actively recommends the by-path name, because /dev/dri/cardN numbering is assigned dynamically at boot and changes. If aquamarine logsdrm: Explicit device <path> not found, the path is wrong or the card really is absent, not merely symlinked. - **Wrong context.** Launching Hyprland from inside an existing X11 or Wayland session, which the error message calls out directly.
Fix
**1. Read the log first. The specific error decides the path.**
cat "$XDG_RUNTIME_DIR/hypr/$HYPRLAND_INSTANCE_SIGNATURE/hyprland.log" 2>/dev/null \
|| cat ~/.local/share/hyprland/hyprland.log
**2. libseat: failed to open a seat. Fix the session.**
loginctl
loginctl show-session "$(loginctl | awk -v u="$USER" '$3==u{print $1; exit}')" -p Type -p Active -p Seat -p Remote
echo "$XDG_RUNTIME_DIR" # must be /run/user/$(id -u), not empty
# Arch/Omarchy: a polkit package OR seatd is mandatory
pacman -Qs polkit
sudo pacman -S polkit hyprpolkitagent # preferred on Hyprland
# alternative: the seatd backend
sudo pacman -S seatd
sudo systemctl enable --now seatd.service
sudo usermod -aG seat,video,input "$USER" # seatd's own group, created by the package
# log out fully and log back in - group changes need a new login session
Then launch from a real virtual console: Ctrl+Alt+F3, log in, and run Hyprland (or uwsm start hyprland.desktop). Never sudo Hyprland.
**3. Found no gpus to use. Fix the GPU side.**
ls -l /dev/dri/ # expect cardN and renderD128
lspci -k -d ::03xx # "Kernel driver in use:" must be set
sudo dmesg | grep -iE 'drm|i915|xe |amdgpu|nouveau|nvidia' | head -40
grep -o nomodeset /proc/cmdline # if this prints, remove it from the cmdline
echo "AQ_DRM_DEVICES=$AQ_DRM_DEVICES" # if set, must be a ':'-separated list of real nodes
grep -i 'Explicit device' "$XDG_RUNTIME_DIR/hypr/"*/hyprland.log
If /dev/dri/ is empty, the kernel driver never bound: a driver problem, not a Hyprland problem (blacklisted module, nomodeset, or hardware needing force_probe).
If the log says drm: Explicit device <path> not found, your AQ_DRM_DEVICES entry is simply wrong. Do **not** "fix" it by swapping a by-path symlink for /dev/dri/cardN. Aquamarine canonicalises both sides, so symlinks work, and the Multi-GPU wiki warns that cardN numbering changes at boot. Use the stable by-path name:
ls -l /dev/dri/by-path/ # match the PCI address from lspci
Where you set it depends on how the session starts. **On Omarchy the session is started by uwsm, and the Hyprland wiki says uwsm users must export it in ~/.config/uwsm/env-hyprland:**
# ~/.config/uwsm/env-hyprland
export AQ_DRM_DEVICES=/dev/dri/by-path/pci-0000:06:00.0-card:/dev/dri/by-path/pci-0000:01:00.0-card
Only if you are *not* using uwsm does the config form apply:
-- ~/.config/hypr/hyprland.lua
hl.env("AQ_DRM_DEVICES", "/dev/dri/by-path/pci-0000:06:00.0-card")
**4. Prove the stack works independently of Hyprland.** The package is drm-info. The binary is drm_info:
sudo pacman -S drm-info && drm_info | head -30
Verify
Hyprland starts from a TTY. hyprctl monitors lists your outputs. loginctl show-session ... -p Active reports Active=yes, and the log contains drm: Found N GPUs instead of the errors above.
Sources
- https://wiki.archlinux.org/title/Hyprland
- https://github.com/hyprwm/Hyprland/blob/main/src/Compositor.cpp
- https://github.com/hyprwm/aquamarine/blob/main/src/backend/Session.cpp
- https://github.com/hyprwm/aquamarine/blob/main/src/backend/drm/DRM.cpp
- https://gitlab.archlinux.org/archlinux/packaging/packages/seatd/-/raw/main/PKGBUILD
- https://github.com/hyprwm/Hyprland/issues/1185