Raspberry Pi 5 performance

[bennypowers]

Hello friends,

I've installed niri on this setup, with noctalia shell:

▓▓▓▓▓▓▓▓ Hardware ▓▓▓▓▓▓▓▓                                -/oyddmdhs+:.
Board: 5-model-b                                      -odNMMMMMMMMNNmhy+-`
CPU: BCM2712 (4) @ 2.40 GHz                         -yNMMMMMMMMMMMNNNmmdhy+-
GPU: Broadcom bcm2712-vc6 [Integrated]            `omMMMMMMMMMMMMNmdmmmmddhhy/`
Display (FHD): 11" FHD 1920x1280                  omMMMMMMMMMMMNhhyyyohmdddhhhdo`
Memory: 921.80 MiB / 3.94 GiB (23%)              .ydMMMMMMMMMMdhs++so/smdddhhhhdm+`
Swap: 0 B / 22.35 GiB (0%)                        oyhdmNMMMMMMMNdyooydmddddhhhhyhNd.
Disk (/): 15.51 GiB / 209.60 GiB (7%) - btrfs      :oyhhdNNMMMMMMMNNNmmdddhhhhhyymMh
Keyboard: Toucan Keyboard                            .:+sydNMMMMMNNNmmmdddhhhhhhmMmy
Mouse: Toucan Mouse                                     /mMMMMMMNNNmmmdddhhhhhmMNhs:
                                                     `oNMMMMMMMNNNmmmddddhhdmMNhs+`
▓▓▓▓▓▓▓▓▓ System ▓▓▓▓▓▓▓▓▓                         `sNMMMMMMMMNNNmmmdddddmNMmhs/.
OS: Gentoo Linux aarch64                          /NMMMMMMMMNNNNmmmdddmNMNdso:`
Kernel: Linux 6.12.75-p20260408-raspberrypi      +MMMMMMMNNNNNmmmmdmNMNdso/-
Uptime: 11 mins                                  yMMNNNNNNNmmmmmNNMmhs+/-`
Packages: 691 (emerge)                           /hMMNNNNNNNNMNdhs++/-`
Shell: fish 3.7.1                                `/ohdmmddhys+++/:.`
                                                   `-//////:--.
▓▓▓▓▓▓▓▓ Desktop ▓▓▓▓▓▓▓▓▓
Terminal: /dev/pts/0
Editor: nvim 0.12.2

Niri starts up just fine thanks to #151, however, when enabling the toggle-debug-tint, I see that all surfaces are always tinted green, regardless of display resolution, scale, or animations being on or off.

On startup, niri logs:

WARN niri::backend::tty: error adding primary node device, display-only devices
may not work: DRM access error: Error loading resource handles on device
`Some("/dev/dri/card0")` (Operation not supported (os error 95))

It appears that card0 fails because vc4 on RPi5 doesn't support drmModeGetResources (display resources are on card1). niri falls back to card1 for display and renderD128 for rendering, which works for basic compositing but not direct scanout.

GNOME/mutter apparently doesn't have this issue, owing to the use of Mesa's KMSRO layer

I'm opening this as a discussion, to see if anyone has workarounds, and since there are related issues already open:

Related issues

• niri Split render/primary devices don't work properly #151 (closed): split render/primary devices -- fixed basic startup, but direct scanout was not addressed
• Smithay Expose libinput Button Scrolling Button Lock Enabled property #1844 (open): "Correct dmabuf feedback configuration for split renderer" -- same class of issue on PinePhone (Allwinner sun4i, also split render/display)
• raspberrypi/linux #6867: VC6 doesn't expose scanout framebuffer via standard DRM APIs

[kmf]

Wonder if its related to the Graphics drivers on the Pi?

[Sempyos]

Perhaps, I would try looking at sway and cosmic-comp to see how they perform

[bennypowers]

swayfx does alright on rpi5

[Sempyos]

Could you do cosmic-comp..?

[bennypowers]

Pop_OS 24.04 LTS doesn't boot on RPi5, but the raw.xz image of fedora cosmic 44 for aarch does :D. I had to manually provision a user from my workstation via a systemd-nspawn root, kind of a fascinating process actually, but after that i finally booted into a graphical session. however i wasn't able to get fedora to load the v3d or vc4 drivers, so in the end it was kind of a waste of time :/

[Sempyos]

Welp, idk, testing another Smithay based compositor (cosmic but that didn't work) would be the easiest way to help tell performance

[kmf]

Maybe an Ubuntu 26.04 ... would work ... did a config for a friend running DMS on it with Niri, wasn't super responsive