CPU resoruces difference, exactly the same video played in bullseye (left) and trixie (right):

Bullseye:

• GStreamer 1.22.0
• Chromium 134.0.6998.35 built on Debian GNU/Linux 12 (bookworm)

Video & audio was played flawlessly.

Trixie:

• GStreamer 1.26.2
• Chromium 140.0.7339.127 built on Debian GNU/Linux 13 (trixie)

Video was missing a lot of frames, audio was played flawlessly.

Maybe something changed in gstreamer?

IIRC Trixie marks the first release wherein Debian defaults to both Wayland and Pipewire regardless of WM/DE. Possible there exist components used for migration (e.g. PA -> PW) or default configs (e.g. for GStreamer) changing to recognize that fact between Bookworm and Trixie (or in this case even longer)? Also does this same relative CPU uplift occur between base images as well or only when certain components of neko stack are introduced?

What about 1:1 comparing how much each service consumes with htop in the docker container, so it's possible to see which service requires more CPU?

TODO: Update Docs for chromium:

It seems that chromium is not woking without --no-sandbox. That could cause some security implcation, but it would also mean we don't need cap_add: SYS_ADMIN anymore.

Tested all nvidia images, fixed chromium there and updated docs. This would be ready to merge. Not sure if this upgrade could negatively affect somebody else using neko.