Another Bug Found That Limits GNOME's Performance For Secondary GPU Setups

[phoronix]

Phoronix: Another Bug Found That Limits GNOME's Performance For Secondary GPU Setups

Daniel van Vugt of Canonical's desktop team for Ubuntu Linux has been on a spree recently tackling various GNOME bugs -- often performance issues -- while also continuing to work on the dynamic triple buffering support and other GNOME desktop enhancements. His latest discovery is around finding another performance bottleneck for multi-GPU setups...

https://www.phoronix.com/news/GNOME-Bug-Secondary-Frame-Rates

[Kjell]

Another bottleneck was just discovered, this time in amdgpu: https://gitlab.freedesktop.org/drm/a...6#note_2119750

Disabling the new FIFO GPU scheduling policy with gpu_sched.sched_policy=0 kernel argument results in lower frame latency

[MrCooper]

It's still 30 fps with disabled dynamic max render time on my dual-AMD system, so there's another bottleneck.

I described the likely root cause of dynamic max render time not working as intended and how it could be addressed in https://gitlab.gnome.org/GNOME/mutte...0#note_1861825.

[MastaG]

Well lets hope that both Gnome/Mutter issues for secondary GPU's and the one for the AMD scheduler are all solved (and merged) before the release of Fedora 40 and Gnome 46.
Then we hopefully have fast and responsive desktop experiences in all scenarios (even the ones who take their Intel/Nvidia laptop to work and use docking stations / usb-c hubs).

[patrakov]

They will also have to fix cursor lag. On a Haswell-era Intel GPU, when using Chromium, moving the mouse results in a major lag when crossing areas that contain hyperlinks (i.e. where the pointer shape changes). No such problem on X. No such problem on KDE with Wayland, and that's probably where I will migrate.

[RussianNeuroMancer]

They will also have to fix cursor lag. On a Haswell-era Intel GPU, when using Chromium, moving the mouse results in a major lag when crossing areas that contain hyperlinks (i.e. where the pointer shape changes). No such problem on X. No such problem on KDE with Wayland, and that's probably where I will migrate.

It's should be fixed in Gnome 45.

[patrakov]

It's should be fixed in Gnome 45.

Unfortunately, it isn't. Tried yesterday on Arch Linux.

[MastaG]

Gnome team seem really careful in reviewing and merging fixes.
Especially when it comes down to the core of mutter.
I don't expect any of the fixes to make it before 45 final.
I think the KDE team are a little more relaxed in this regard.

[conualfy]

Just recently they made the cursor to be in a separate thread on Wayland; this caused the lag on slower machines. But it is not yet in production.

It's should be fixed in Gnome 45.