Firefox Nightly Tries For VA-API Video Acceleration For Mesa Users

[oleid]

Here's the only recent performance indirect comparison between GTK3 and GTK4 that I've found:

https://www.reddit.com/r/gnome/comments/tbiw77/scrolling_comparison_between_web_gtk3_vs_gtk_4/

To be honest I see zero difference between software rendered GTK3 and OpenGL accelerated GTK4.

I would be highly surprised if there was a visual difference in this video. The only real difference should be in cpu load. Animations are capped at 60 FPS anyway.

You could try gtk4-demo and check out your process monitor or something.

[mdedetrich]

Are you using a Pentium 4? One of the main issue why GPU based video en/decoding is such a slow starter on Linux is that it became really unnecessary in the meantime for Workstations. Video Rendering with a CPU based pipeline just causes such minimal CPU usage on anything from the last 10 years that there simply is not much demand and low demand means low motivation.

So these are modern work laptops, specifically in the case of my colleague and myself its Thinkpad x1 Carbon Gen-9. So not, its not a Pentium 4 but an i7-1165G7 which has integrated graphics.

This is btw just one of the problem's, in general our company (which is Fedora/Thinkpad based for our procured hardware) has so many issues with Linux desktop that we actually calculated the lost productivity (can't say how much it is, but its massive) which is why along with other reasons (supply chain issues) we started giving option for people getting Macbook Pro m1. And to put things into context since the Macbook Macbook Pro M1 has ARM architecture, its still cheaper to spend programmer time for us to port our software to ARM so that it can be run on the Apple M1 developer machines.

[user1]

VA-API was designed to be windowing system and hw agnostic.

Hardware agnostic, but Mesa exclusive. So it being hardware agnostic doesn't really help in that case.

[birdie]

I would be highly surprised if there was a visual difference in this video. The only real difference should be in cpu load. Animations are capped at 60 FPS anyway.

You could try gtk4-demo and check out your process monitor or something.

gtk4-demo by itself is not a CPU intentsive application. gtkperf is but again no one has bothered to port it to GTK3/4.

[birdie]

Because until recently the GPU acceleration of the shell wasn't very efficient. Performance fixes are in GNOME 42.

OMG. Year 2022 and Linux still renders UI in software mode.

[birdie]

VDPAU is Xlib only, it doesn't even support Xcb, let alone Wayland. It did its job at the time of getting accelerated video working on X11 with NVidia GPUs, but it wasn't really designed with anything else in mind. VA-API was designed to be windowing system and hw agnostic.

As far as I can read C code VDPAU barely uses anything X11 specific: https://gitlab.freedesktop.org/vdpau/libvdpau

There's very little code and it can be trivially converted to another graphical system, including Wayland.

[Ermine]

How can I be sure about that? Is there a simple test to run to verify that? A demo or something? Also, is it just GTK4? What about GTK3? Qt5? Qt6?

Again, I need solid proofs, not "I've heard about Vulkan and OpenGL backends" - I did as well. Doesn't make me all so confident.

There are quite annoying bug reports on the topic: https://gitlab.gnome.org/GNOME/gtk/-/issues/4112

Qt5/6 have examples of opengl/vulkan apps. Go check them.

[Quackdoc]

Hardware agnostic, but Mesa exclusive. So it being hardware agnostic doesn't really help in that case.

it's not mesa exclusive. Nvidia has a vaapi implementation, v4l2 has a vaapi implementation. and I think someone was looking at maybe implementing a vaapi ontop of vulkan video too

[Ermine]

OMG. Year 2022 and Linux still renders UI in software mode.

This is non sequitur.

[Ermine]

it's not mesa exclusive. Nvidia has a vaapi implementation, v4l2 has a vaapi implementation. and I think someone was looking at maybe implementing a vaapi ontop of vulkan video too

Nvidia does not implement VA-API. This one is unofficial.