Vkcube: ~300 more FPS on XWayland vs Wayland

Hello there,

I have noticed that, on my system, using XWayland for Vulkan & OpenGL applications results on my dedicated NVIDIA GPU in much better performance than using Wayland. This also applies for DXVK & VKD3D (which means it affects games I play through Proton as well as native apps). I have reproduced the issue with vkcube (using the Xcb backend for XWayland). Here are my system specifications:

CPU: Intel Core i5-11400H

RAM: 2x8 GB DDR4 3200 MHz

GPU: NVIDIA GeForce RTX 3050 Laptop GPU

Driver version: 610.43.03

Operating system: Arch Linux

Kernel version: 7.1.3-arch1-2

Desktop environment: KDE Plasma 6.7.2

I have used the following commands to run vkcube:

MANGOHUD=1 MANGOHUD_CONFIG=test.conf vkcube --wsi xcb --present_mode 0
MANGOHUD=1 MANGOHUD_CONFIG=test.conf vkcube --wsi wayland --present_mode 0

Where test.conf just contains the output_folderoption for MangoHud so I can log the results. And speaking of logs, here are they:

XWayland:

vkcube_2026-07-10_17-23-50_summary.zip (405 Bytes)

Wayland:

vkcube_2026-07-10_17-24-20_summary.zip (406 Bytes)

I do have the complete MangoHud logs if that’s necessary. Here are screenshots that show the performance gap in real-time (first one is Wayland, second one is XWayland):

Finally, I have created two bug reports using nvidia-bug-report.sh, one for each case (XWayland and Wayland vkcube WSIs) when running vkcube:

nvidia-bug-report-wayland.log.gz (1.4 MB)

nvidia-bug-report-xwayland.log.gz (1.4 MB)

Although, the tool complained that mst and nvlsm-bug-report.sh weren’t present, I’m not sure if that’s important.

Finally, I’m happy to share any more information needed!

Side note: I have the KWIN_USE_OVERLAYS=1 environment variable set here (which I believe is set by default now). When setting it to 0, the GPU usage (and thus the FPS) spike up considerably, but there is still a wide performance gap between XWayland (~6000 FPS) vs. Wayland (~3000 FPS).