https://bugs.kde.org/show_bug.cgi?id=522041
Bug ID: 522041
Summary: KWin screencast of 4K@240Hz monitor causes severe CPU
load and system sluggishness (Plasma 6.7 regression,
NVIDIA Wayland)
Classification: Plasma
Product: kwin
Version First 6.7.0
Reported In:
Platform: Other
OS: Linux
Status: REPORTED
Severity: normal
Priority: NOR
Component: screencasting
Assignee: [email protected]
Reporter: [email protected]
Target Milestone: ---
SUMMARY
-------
After upgrading to Plasma 6.7 on Fedora 44, sharing my primary 4K monitor
through the xdg-desktop-portal-kde screen share picker causes severe
system-wide sluggishness. Sharing a single application window on the same
monitor, or sharing either of my two smaller 1080p portrait monitors, does
not cause the same problem.
Performance profiling shows KWin (kwin_wayland) CPU usage increases
significantly when capturing the full 4K output vs. a single window, with
combined KWin + client CPU exceeding 200% during full-monitor capture.
This did not appear to be an issue before upgrading to Plasma 6.7. The portal
picker UI also changed in 6.7, adding "Share full Workspace", "Share virtual
screen", and "Share region" options.
STEPS TO REPRODUCE
------------------
1. Boot into KDE Plasma Wayland session
2. Open Discord (or any app requesting screen share via xdg-desktop-portal)
3. Click "Share Screen"
4. In the KDE portal dialog, select "Share 'LG Electronics LG ULTRAGEAR+
(DP-3)'"
5. Start the stream
6. Observe: severe system sluggishness, low desktop FPS, unresponsive UI
For comparison (no sluggishness):
(a) Select a single application window on DP-3 instead of the full monitor
(b) Select either 1080p portrait monitor (DP-1 or DP-2)
EXPECTED RESULT
---------------
Sharing a single monitor should not cause severe system-wide performance
degradation on an RTX 3070 system.
ACTUAL RESULT
-------------
Full DP-3 monitor share causes severe sluggishness. Window share on the same
monitor and full share of smaller monitors work acceptably.
PERFORMANCE DATA
----------------
Captured via: top -b -n 3 -d 2 | grep -iE 'kwin|pipewire|discord|plasmashell'
A) Full DP-3 monitor share (PROBLEMATIC):
Discord (PID 36245): 140–149% CPU
kwin_wayland (PID 11916): 66–71% CPU
pipewire (PID 1645): 0–1.5% CPU
Combined: ~210%+ CPU
B) Single window share on DP-3 (ACCEPTABLE):
Discord (PID 36245): 127–129% CPU
kwin_wayland (PID 11916): 42–54% CPU
pipewire (PID 1645): 0–1.5% CPU
Combined: ~175% CPU
Delta (full monitor vs window):
kwin_wayland: +20–25% CPU (primary KDE-side cost increase)
Discord: +15–20% CPU
pipewire: no meaningful difference (not the bottleneck)
The KWin screencast overhead for full 4K@240Hz output capture appears to be
the primary compositor-side regression. PipeWire itself consumes negligible CPU
in both scenarios.
REGRESSION
----------
Appears introduced with Plasma 6.7 upgrade. Not observed on Plasma 6.6.x.
Packages updated around when issue started:
plasma-workspace-6.7.0-1.fc44 (2026-06-17)
kwin-6.7.0-1.fc44 (2026-06-17)
xdg-desktop-portal-kde-6.7.0-1.fc44 (2026-06-17)
xdg-desktop-portal-1.22.1-1.fc44 (2026-06-22)
pipewire-1.6.7-1.fc44 (2026-06-20)
ENVIRONMENT
-----------
OS: Fedora 44 (Forty Four)
Kernel: 7.0.12-201.fc44.x86_64
Session: Wayland
GPU: NVIDIA GeForce RTX 3070, driver 595.80 (proprietary, akmod)
KWin: 6.7.0
Plasma: 6.7.0
Portal: xdg-desktop-portal 1.22.1, xdg-desktop-portal-kde 6.7.0
PipeWire: 1.6.7
Discord: 1.0.143 (native RPM)
Display setup (3 monitors):
DP-3: LG Electronics LG ULTRAGEAR+
Native 3840x2160 @ 240.08 Hz, logical 2560x1440, scale 1.5, VRR:
Automatic
DP-1: LG Electronics 27GL650F
1920x1080 @ 143.98 Hz, portrait, scale 1.0, VRR: Never
DP-2: LG Electronics 27GL650F
1920x1080 @ 143.98 Hz, portrait, scale 1.0, VRR: Never
WORKAROUNDS
-----------
- Share individual application windows instead of full monitor
- Share DP-1 or DP-2 portrait monitors (no issue)
- KWIN_DRM_NO_DIRECT_SCANOUT=1 not tested permanently (concern about fullscreen
game latency impact)
ADDITIONAL NOTES
----------------
- Issue is reproducible via Discord but likely affects any portal-based
full-output screencast
- journalctl shows repeated plasmashell "Failed to connect to PipeWire" errors,
though pipewire service is running and shows negligible CPU during capture
- New Plasma 6.7 portal options (Workspace / virtual screen / region) coincide
with when issue was first noticed
--
You are receiving this mail because:
You are watching all bug changes.
