Firefox 80 To Support VA-API Acceleration On X11

[caligula]

Using the gpu is not hardware video decoding, the hardware video decoding is another matter. Every video card has a separate integrated part for hardware decoding the (supported) video streams.

It doesn't matter if you decode using GPGPU cores or DSP, the process node advancements still apply. I'm just saying super low power devices could decode H264 already 8 years ago. Now I have a fairly recent 14nm Intel chipset and the CPU load is around 70% (single core) when playing 720p H264 in Firefox. I'm pretty sure the $1000 laptop is better than first gen RPi in all possible ways. Still the power consumption is much higher when watching Youtube.

[ms178]

As a user, what do you care about the market share? Why would you choose a browser based on its market share?

It is quite a good indicator how successful a browser is - the market share is at that point for several reasons for which performance and overall usability are two important factors where Firefox just is lacking behind. I switched to Chrome during the first few weeks after release, it was just faster and on my notebook it mattered a lot. Hardware video decoding is an important feature to me though, and I would need to re-evaluate my options if Chromium upstream remains hostile to include it.

[nanonyme]

Impossibru. According to a lot of Phoronix members, this would be a waste of time and resources and Mozilla wasn't working on this. So is this news real?

Someone said it was actually Red Hat that contributed this in the end.

[ernstp]

Gnome's Web/Epiphany actually has hardware decoding for a few years now.

True!

[molecule-eye]

Not really. If you cared about power saving, you'd extract the video URL and just use MPV. I tried enabling the VA-API on Wayland & FF78 and the performance is still bad. I'd expect h264 to consume almost zero power since even the original single core 700 MHz RPi could play 1080p video without any issues. First 1080p video capable dual-core tablets and phones appeared around 8 years ago. With each new CPU generation they advertise how it consumes 50% less power while providing 50% more computational power. So basically the 8 year old 10W devices could do this, now you have 2**8 = 256 times better power efficiency.

This is what bugs me. I have old ARM hardware that runs 1080p60fps no problem, without dropping a frame, and yet performance on linux in a browser--on well supported hardware!--is dismal. This is why I still prefer using my old, fanless, cool, and quiet and underpowered tablet in bed over my way faster, more convenient, fan-spinning laptop for watching HD videos.

[ernstp]

Not really. If you cared about power saving, you'd extract the video URL and just use MPV. I tried enabling the VA-API on Wayland & FF78 and the performance is still bad. I'd expect h264 to consume almost zero power since even the original single core 700 MHz RPi could play 1080p video without any issues. First 1080p video capable dual-core tablets and phones appeared around 8 years ago. With each new CPU generation they advertise how it consumes 50% less power while providing 50% more computational power. So basically the 8 year old 10W devices could do this, now you have 2**8 = 256 times better power efficiency.

Browsers have to convert the Video from YUV or whatever they get it in to RGB before displaying it, to be able to combine with media controls and other things they have. Native video players generally don't have to do that so the save a lot of processing.

[brent]

Browsers have to convert the Video from YUV or whatever they get it in to RGB before displaying it, to be able to combine with media controls and other things they have. Native video players generally don't have to do that so the save a lot of processing.

This is pretty much completely wrong. Standalone video players need to convert to RGB too, of course. Otherwise they wouldn't be able to display it on your RGB display. And you also have OSD and subtitles in standalone video players, and these elements need to be composited with the video content. Browsers and standalone video players aren't actually very different at all.

Firefox' implementation of accelerated video playback uses VAAPI for decoding and OpenGL shaders for presentation and compositing, so in theory it should be just as efficient as mpv (< 10% CPU load), but it seems to have a ton of CPU overhead for whatever reason. It needs a lot of CPU cycles for things not strictly related to video decoding or presentation. Let's hope they'll sort that out.

[bug77]

Browsers have to convert the Video from YUV or whatever they get it in to RGB before displaying it, to be able to combine with media controls and other things they have. Native video players generally don't have to do that so the save a lot of processing.

This is what the conversion looks like: https://en.wikipedia.org/wiki/YUV#HDTV_with_BT.709
There's no way 9 multiplications and 6 additions need that much extra CPU time. Though back in the day, I remember Adobe complaining about YUV as well...

[240Hz]

But xOrG iS abAndoNEd

144Hz

[bug77]

But xOrG iS abAndoNEd

144Hz

Please don't. It's strange enough he stayed away so far, do you dislike 5 pages of readable posts that much?