You are not logged in.
Anyone else also seeing this? I only have one machine with a video card without shared memory (is there a tool that tells you what's using VRAM for AMD/Intel GPUs?), but on my nvidia machine, on kwin_wayland and hyprland, I'm seeing Xwayland use up to 2.4GB of VRAM and not letting it go, even after I kill all the X11 apps.
I've filed a bug here:
https://gitlab.archlinux.org/archlinux/ … -/issues/1
Anyone else seeing this on say an AMD machine? or perhaps other compositors?
Offline
My AMD hardware does not exhibit this behaviour. Checked with xterm & nvtop under sway.
Jin, Jîyan, Azadî
Offline
nvidia is notoriously slow in re-allocating GL contexts, ie. eg. resizing them.
Maybe the driver side-steps that by allocating the maximum buffer.
Can you re-claim the VRAM by
1. terminating the xwayland server (not only the clients) server
2. and then running a VRAM intense process?
Offline
My AMD hardware does not exhibit this behaviour. Checked with xterm & nvtop under sway.
Strangely this doesn't happen with xterm. Try launching firefox with:
env WAYLAND_DISPLAY="" firefoxFor some reason this doesn't happen with xterm.
I've tested this on my framework laptop (which uses an APU, though), and I don't see the problem in nvtop there.
Last edited by kelvie (2023-12-29 23:32:37)
Offline
nvidia is notoriously slow in re-allocating GL contexts, ie. eg. resizing them.
Maybe the driver side-steps that by allocating the maximum buffer.Can you re-claim the VRAM by
1. terminating the xwayland server (not only the clients) server
2. and then running a VRAM intense process?
Terminating the xwayland server kills the clients, so yes that works, it requires a -9 though, it doesn't respond to SIGTERM.
Right now I have a workaround that kills all xwayland programs (gracefully) then force killing xwayland.
Offline