https://bugs.kde.org/show_bug.cgi?id=527750
Bug ID: 527750
Summary: Fullscreen games present at exactly half the refresh
rate when per-display color preference is "Prefer
efficiency" (regression in Plasma 6.8.0)
Classification: Plasma
Product: kwin
Version First 6.7.5
Reported In:
Platform: Arch Linux
OS: Linux
Status: REPORTED
Severity: normal
Priority: NOR
Component: wayland-generic
Assignee: [email protected]
Reporter: [email protected]
Target Milestone: ---
Created attachment 197615
--> https://bugs.kde.org/attachment.cgi?id=197615&action=edit
Output for `qdbus org.kde.KWin /KWin supportInformation`
Version: 6.8.1 (regression introduced in 6.8.0)
Platform: Arch Linux, kwin_wayland session
Environment:
• Kernel 6.18.58-1-lts
• AMD Radeon RX 9060 XT (Navi 44), mesa 26.2.5, vulkan-radeon 26.2.5 (RADV)
• Display: 2560x1440 @ 59.95 Hz, VRR incapable, kwin VRR policy "Never"
• Game used for reproduction: Mewgenics (Steam AppID 686060) via Proton 9.0
(Beta), native Vulkan renderer through winevulkan, running fullscreen on
Xwayland; MangoHud overlay for measurements
Steps to reproduce:
1. On a Plasma 6.8.x Wayland session with a display set to the default "Prefer
efficiency" color preference [adjust to the exact label you saw in the UI],
launch a vsynced fullscreen Vulkan game on a 60 Hz display.
2. Arrange for the game to render some per-frame content that pushes frame
time just over one refresh period (in my case: hovering a map tile renders an
extra highlight pass). Content that renders under budget does not trigger it.
3. Observe the framerate.
Actual results:
• The game presents at exactly half the refresh rate: 30 fps on a 60 Hz
display. The halving is consistent, not stuttery — one vblank is missed per
frame, and with FIFO present the queue settles on every other vblank.
• Neither CPU (~25% of one core) nor GPU (~15% busy) is saturated during the
drop; GPU clocks fall back and wander between 600-900 MHz, indicating the
presentation path is blocked rather than the hardware being loaded.
• With in-game vsync disabled, the game still submits ~60 fps (per MangoHud)
but on-screen motion remains visibly slowed, indicating submitted frames are
delivered to scanout late.
• The problem does not occur in windowed mode, and does not occur in menus or
whenever per-frame render cost stays comfortably below one refresh period.
Expected results: 60 Hz presentation in fullscreen, matching windowed behavior.
Workaround: Switching the display's color preference from "Prefer efficiency"
to "Prefer accuracy" restores correct fullscreen presentation (60 fps). This
points at the efficiency-path scanout/color handling introduced in 6.8 losing
vblanks under marginal per-frame loads.
Ruled out:
• Mesa/RADV: reproduced identically on mesa 26.2.5 and 26.2.4 (downgrade test)
• GPU clocks/power management: forcing DPM level to high has no effect
• VRR: display is VRR-incapable, policy "Never"
• Game update: game binary unchanged since before the regression appeared
• Kernel logs are clean (no amdgpu resets/timeouts)
Regression window: Plasma 6.7.5 → 6.8.0, first observed 2026-10-21, still
present in 6.8.1.
Related reports: Same signature (fullscreen choppy/low fps, windowed fine, AMD
GPUs incl. RX 6700 XT / 7600 / 9070 XT / 9060 XT) reported by multiple users
since Plasma 6.8.x, with "Prefer accuracy" or downgrade to 6.8.0 as
workarounds:
https://discuss.cachyos.org/t/plasma-6-8-1-update-fullscreen-games-choppy-fine-in-windowed-desktop-animations-laggy-fixed-by-downgrading-to-6-8-0/38095
--
You are receiving this mail because:
You are watching all bug changes.
