NVIDIA on Fedora KDE Wayland in 2026: Field Report from an RTX 5070

· 5 min read

NVIDIA on Linux has been a punchline for so long that “just buy AMD” was the standard advice in every Linux subreddit thread. In 2026, on Fedora 44 with KDE Plasma 6.6 on Wayland, running an RTX 5070 (Blackwell) with the 595 driver branch, I want to say something different:

It’s fine. It’s actually fine.

Not “fine if you don’t use Wayland.” Not “fine if you stick to X11.” Fine, full stop. This is the short, honest report of how I run it.

Where we were

The historical pain, in one paragraph: NVIDIA shipped its own EGL implementation, refused to ship GBM, refused to ship implicit sync the way Mesa wanted it. Wayland compositors had to choose between writing NVIDIA-specific code paths (EGLStreams) or telling NVIDIA users to use X11. KDE did the latter for years. The result was the meme: Plasma on Wayland with NVIDIA? Don’t.

What actually changed

Four things, in order of impact:

  1. NVIDIA started shipping GBM (driver 495 in 2021), bringing the buffer-allocation model in line with everyone else. EGLStreams stopped being the only path.
  2. GSP firmware moved the driver’s brain off the host kernel and onto the GPU itself. Turing and later GPUs run a chunk of the driver inside a dedicated microcontroller (the GPU System Processor). The open-source kernel modules became viable because what runs in the kernel is now thin.
  3. The open kernel modules became default and then mandatory. For Blackwell (RTX 50 series), the open modules are the only option — the proprietary kernel modules don’t support these GPUs at all.
  4. Explicit sync landed in Wayland and in KWin. The compositor and the driver now agree on when a buffer is ready to be displayed, removing a category of glitches that affected NVIDIA on Wayland specifically.

Put those four together and the result is what I’m typing this post on.

The install, on Fedora 44

Two commands. That’s the whole thing.

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

sudo dnf install akmod-nvidia

That’s it. akmod-nvidia pulls in the GSP firmware, the userspace libraries, nvidia-modprobe, the Xorg driver (for legacy apps), and the kernel module sources. The akmods service then compiles the module in the background. Wait 2–5 minutes before rebooting — if you reboot during the build, you boot into Nouveau and assume the install failed. It didn’t; you just rushed it.

Why akmod and not kmod or DKMS

Three flavors exist; only one of them is right on Fedora:

  • kmod-nvidia — a precompiled module for a specific kernel version. Works until the next kernel update, then breaks until someone rebuilds it.
  • DKMS — the Ubuntu/Debian-world solution. Rebuilds on kernel update via DKMS hooks.
  • akmod-nvidia — RPMFusion’s equivalent of DKMS, but tuned for Fedora’s RPM hooks and kernel package layout. Rebuilds the module automatically whenever a new kernel lands. This is the one.

On my system, rpm -qa | grep kmod-nvidia shows three compiled modules sitting side by side — one for each of the last three kernels I’ve had installed:

kmod-nvidia-7.0.7-200.fc44.x86_64-595.71.05-1.fc44
kmod-nvidia-7.0.8-200.fc44.x86_64-595.71.05-1.fc44
kmod-nvidia-7.0.9-202.fc44.x86_64-595.71.05-1.fc44

Old modules get garbage-collected as old kernels are removed. I never touched any of this manually. That’s the whole point.

What to check after install

Five quick checks I run on a fresh install:

# 1. Module is built and loaded
modinfo -F version nvidia
# → 595.71.05

# 2. nvidia-drm.modeset is enabled (required for Wayland)
cat /sys/module/nvidia_drm/parameters/modeset
# → Y

# 3. Open kernel module variant in use (required for Blackwell)
modinfo nvidia | grep -i license
# → license:  Dual MIT/GPL

# 4. Wayland session, not X11
echo $XDG_SESSION_TYPE
# → wayland

# 5. GPU is visible and driver matches running kernel
nvidia-smi --query-gpu=name,driver_version --format=csv,noheader
# → NVIDIA GeForce RTX 5070, 595.71.05

If any of those five lines disagrees, that’s where the problem is. The rest is application-level.

What still hurts, honestly

Three things I’m not going to pretend are perfect:

  1. Suspend-to-idle on kernel 7.0 with Blackwell can hang on resume. There’s an open issue against the open kernel module repo (#1117). The 6.17 kernel works around it; 7.0 reintroduced something that affects RTX 50 specifically. Workaround for now: s2idle instead of deep in /sys/power/mem_sleep, or skip suspend entirely on desktops where it doesn’t matter.
  2. VRR (variable refresh rate) on some monitors needs manual configuration through KScreen-doctor or by editing the KWin per-output config directly. The default “auto-detect VRR” still misses some panels.
  3. Secure Boot adds a step. The akmod-built module isn’t signed by default. Either disable Secure Boot, or generate a MOK (Machine Owner Key), enroll it, and configure akmods to sign with it. The RPMFusion wiki covers this; the one-line summary is kmodgenca -a && mokutil --import /etc/pki/akmods/certs/public_key.der.

Everything else — gaming on Wayland, OBS screen capture via PipeWire, multi-monitor with different refresh rates, HDR on a compatible display — works without thinking about it.

Closing

NVIDIA on Linux stopped being a topic worth complaining about somewhere between the 555 and 595 driver branches. On Fedora, with RPMFusion’s akmod-nvidia, the maintenance cost is zero — kernel updates rebuild the module silently, driver updates come through the same dnf upgrade that brings the rest of the system.

If you’ve been avoiding NVIDIA on Wayland because of what was true in 2022, it’s worth a fresh look in 2026. Blackwell on Plasma Wayland genuinely just works.