Firefox 76 Released With WebRender Improvements, Better Security

[pal666]

use brave browser, is working vaapi with h264

i doubt brave supports anything not supported by chromium. i don't have vaapi, but vdpau h264 works

[M@GOid]

If your system gets loud under the load of just decoding video, I'd highly recommend thoroughly cleaning you fans/heatsink and installing fresh thermal compound. I had an old laptop (I think it was of those old 3-core AMD precessors) that served as a media center PC for a while, but when I first set it up, the fan noise was unbearable. After applying some decent thermal compound to the CPU and chipset, it was SIGNIFICANTLY quieter.

You can buy good thermal compound online for like $10, and one tube has enough for multiple systems (especially laptops, that don't need much due to the small die area).

I had laptops over my table that were barely 5 years old and had the heat-sink completely obstructed. I even took videos of it to show the owners why their once snappy PC was slowing down so much.

[andre30correia]

i doubt brave supports anything not supported by chromium. i don't have vaapi, but vdpau h264 works

you doubt? try it them is quiet simply, vaapi is working vdpau is only for nvidia graphs to render flash or someting, last time I check chromium is only the base you can change anything, try it

[lakerssuperman]

I'm having problems with the Flatpak version and video. When I try to play Youtube videos they are slow and jerky. All the Wayland and VAAPI bits have been enabled, so I'm not sure what the issue is. I'm on Fedora and Firefox 75 in the repos works without this issue. I haven't found anything to fix this issue so far, but it only came out today so hopefully something turns up.

[Gusar]

I'm not even sure why video hardware acceleration has anything to do with the display server. I mean, sure, the display server has to present the result, but why does it care where and how the result comes from?

If one uses a toolikt like GTK or Qt, display server interaction ins mostly done for you by the toolkit. But Firefox is effectively it's own toolkit (it does use GTK, but only to inherit its looks), so the devs need to code display server interaction themselves. In the case of hardware video surface handling, the devs have done the interaction with Wayland, but not with X. No technical obstacle, it just hasn't been coded yet. And won't be by the dev who coded the Wayland part, someone else will need to do it.

Now to all those people saying it's impossible on X, that X supposedly doesn't support dma-buf, or that in X one needs to do copy-back. I don't know where these statements come from, or why do they get repeated all the time, because they're completely wrong. You can do dma-buf in X, you don't need the performance-destroying copy-back. Players like mpv are proof of that.

There is one difference between Wayland and X. In Wayland, you can delegate the video rendering stage (note, *not* the decoding, but the rendering) to the compositor by using subsurfaces, in X you need to do render yourself. But Firefox already has a hardware renderer called Webrender, so they could hook up rendering dma-buf surfaces with that. It just needs, as I wrote above, someone to actually code it.

[kylew77]

Good thing I live somewhere where watching a video in anything higher than 360p causes it to buffer cause of the low speed Internet rurally here. Have no need for GPU accelerated video decoding since my processors even in my weakest laptops can handle 360p just fine.

[AnAccount]

Good question!
I was thinking myself that decoding a video should be as simple as passing the video stream to ffmpeg, which will decode it using the CPU or the GPU and give back the result.
The the result should be displayed the same way as it is now when you play some video.
I don't get why could it not be so simple.

The thing is that copying content to and from the GPU is expensive, so sending the data to the GPU, having it rendered, send it back to compose it and then send it back to the hardware to display it will be more inefficient than just render it in CPU. This is how hardware works, and have nothing to to with firefox overcomplicating things. Wayland is designed from scratch with the idea of letting applications have access to the hardware buffers for rendering, making it easier for firefox to render a webpage in the hardware buffer, then letting the video be decoded straight in the hardware buffer and then do the compositing in side the hardware buffer, hence eliminating any extra copies to and from the hardware buffers.
And while this is also possible to do in X, you must remember that X was not designed for giving programs access to the hardware buffers, quite the opposite. So the whole thing with this is an afterthought using extensions. To that you must also remember that firefox X support was designed a long time ago, not necessary with this use case in mind. So it might be quite complicated to rewrite firefox X support to avoid these extra copies that would void the benefits. This is in fact one of the reasons the Xorg developers started Wayland, to create a platform that is more in-line with modern hardware and hence making it easier to take advantage of the hardware.

[Guest]

Any progress on support hardware video decoding on X?

This kind of hardware acceleration is a fake hardware acceleration. Under Xorg hardware acceleration can be implemented fully, if the gpu support hardware decoding. Mplayer, Vlc, Mpv, Xine, etc. all support the hardware decoding. Also the Adobe flash plugin support this.

This Firefox and Chromium solution moves some works from the cpu to the gpu. I'm wrong on this point?

[woprandi]

How to check if the hardware decoding is enabled and works ? I felt is was already enabled on Youtube by watching the %CPU but in videoconference not....

[Tsih]

How to check if the hardware decoding is enabled and works ? I felt is was already enabled on Youtube by watching the %CPU but in videoconference not....

https://wiki.archlinux.org/index.php/Firefox#Hardware_video_acceleration