Sway does not officially support NVIDIA’s proprietary driver. I run it anyway on Fedora — but not with the stock session and not with GDM or SDDM. Both of those gave me a black screen after the proprietary driver was installed: login manager visible, session starts, no picture.

This post documents the setup that actually works on my machine, in the order I applied it.

System and hardware

This guide was written and tested on the following setup. If your GPU or driver branch differs, the overall flow should still apply — but your package names may vary.

Current system:

GPUs and driver branches I tried:

The 580xx packages were specific to the GTX 1060 period. On the RTX 3060 Ti I switched to the rolling akmod-nvidia package, which tracks the latest RPM Fusion branch (595.80 at the time of writing). Kernel parameters, nvidia.conf, and the custom Sway session did not change between the two cards.

1. Enable RPM Fusion (nonfree)

Fedora does not ship the proprietary NVIDIA driver. I only added the nonfree RPM Fusion repo — that was enough for the NVIDIA packages I needed.

From my shell history:

sudo dnf install https://download1.rpmfusion.org/nonfree/fedora/rpmfusion-nonfree-release-$(rpm -E %fedora).noarch.rpm

After that, these repo files showed up under /etc/yum.repos.d/:

• rpmfusion-nonfree.repo
• rpmfusion-nonfree-updates.repo
• rpmfusion-nonfree-updates-testing.repo (disabled by default)

The enabled ones I actually use:

rpmfusion-nonfree
rpmfusion-nonfree-updates

No separate rpmfusion-free repo on my system — nonfree alone covered the driver install.

2. Install the proprietary driver

Once RPM Fusion was in place, on the RTX 3060 Ti I installed:

sudo dnf install akmod-nvidia xorg-x11-drv-nvidia-cuda

On the older GTX 1060 setup I had previously used the 580xx branch instead:

sudo dnf install akmod-nvidia-580xx xorg-x11-drv-nvidia-580xx

Pick the branch that matches your GPU generation in RPM Fusion. The rolling akmod-nvidia package is what I run today on the 3060 Ti (595.80). Reboot. Then confirm the module loaded:

nvidia-smi
lsmod | grep nvidia

If nvidia-smi fails, check /var/log/akmods/ before touching anything Sway-related.

3. Kernel parameters

I did not use a modprobe drop-in. The full GRUB_CMDLINE_LINUX line from /etc/default/grub on my system:

GRUB_CMDLINE_LINUX="rhgb quiet nvidia-drm.modeset=1 rd.driver.blacklist=nouveau,nova_core modprobe.blacklist=nouveau,nova_core"

What each part does:

• nvidia-drm.modeset=1 — kernel mode setting for Wayland
• rd.driver.blacklist=nouveau,nova_core — keep the open-source driver off at early boot
• modprobe.blacklist=nouveau,nova_core — same blacklist once the system is up

After editing /etc/default/grub, regenerate the bootloader config and reboot:

sudo grub2-mkconfig -o /boot/grub2/grub.cfg
sudo reboot

Verify modesetting after boot:

cat /sys/module/nvidia_drm/parameters/modeset

You want Y.

4. NVIDIA environment variables

Sway needs a few wlroots/NVIDIA variables. Mine live in ~/.config/environment.d/nvidia.conf:

WLR_NO_HARDWARE_CURSORS=1
GBM_BACKEND=nvidia-drm
__GLX_VENDOR_LIBRARY_NAME=nvidia

WLR_NO_HARDWARE_CURSORS=1 fixes the invisible cursor. Without it I had a working desktop and no pointer.

5. Why a custom Sway session — and why ly

Stock GDM and SDDM both failed with the proprietary driver: black screen at login, no output at all. The display manager itself was not usable on my setup.

The fix was a TTY-based display manager. I installed ly — minimal, runs on a TTY, no graphical greeter fighting the NVIDIA stack.

Then I created a Sway (Custom) session instead of using the default sway.desktop. Reasons:

• Sway refuses proprietary NVIDIA unless you pass --unsupported-gpu
• I needed a single entry point that sets Wayland env vars and starts Sway correctly
• A custom .desktop file survives package updates better than patching the stock one

Session desktop file

/usr/share/wayland-sessions/sway-custom.desktop:

[Desktop Entry]
Name=Sway (Custom)
Comment=An i3-compatible Wayland compositor
Exec=/home/osman/start-sway.sh
Type=Application

Adjust the Exec= path to wherever you put your startup script.

Startup script

~/start-sway.sh:

#!/bin/bash
# Export Wayland variables
export XDG_CURRENT_DESKTOP=sway
export XDG_SESSION_TYPE=wayland
export MOZ_ENABLE_WAYLAND=1
export CLUTTER_BACKEND=wayland

# Optional: Disable hardware cursor if you experience flickering
# export WLR_NO_HARDWARE_CURSORS=1

# Execute sway inside a dbus-run-session
exec dbus-run-session sway --unsupported-gpu

Make it executable:

chmod +x ~/start-sway.sh

I keep WLR_NO_HARDWARE_CURSORS in ~/.config/environment.d/nvidia.conf instead of the script — one place for NVIDIA-specific vars, script for session bootstrap only.

At the ly login screen, select Sway (Custom) and log in.

6. Verify

Quick checks after login:

echo $GBM_BACKEND
echo $WLR_NO_HARDWARE_CURSORS
nvidia-smi
swaymsg -t get_version

You should see the NVIDIA driver active, environment variables set, and Sway running without the proprietary-driver refusal message.

Common problems

Logs when stuck:

journalctl -b | grep -iE 'nvidia|sway|ly'

Closing notes

This is not an officially supported configuration. Sway upstream and the Fedora Sway spin both distance themselves from proprietary NVIDIA. I document it because it is what runs on my machine — ly on a TTY, custom session, RPM Fusion nonfree, and explicit kernel blacklists for nouveau.

If you want fewer moving parts on Fedora, GNOME or KDE handle NVIDIA with less friction. I wanted Sway, so I accepted the workarounds.