Vulkan in theory should be more flexible, but in practice VAAPI still has some advantages when it comes to power efficiency due to Vulkan path lacking some needed features (I think something to do with color space conversions which in Vulkan now have to happen using regular GPU compute units which is inefficient). In the longer term, Vulkan should be preferable in all scenarios.
Still, isn't this bizarre? Over the last couple of decades using Windows and macOS across many machines, I've encountered one time I had to mess with settings like this. That's the time I temporarily disabled hw accel in chrome so I could screenshare and watch a show with someone.
Meanwhile on Linux you gotta configure every app one at a time because sane defaults and auto configuration weren't invented here.
In addition to what others have said, hardware decoding comes with other drawbacks. There’s occasionally decoding inaccuracies and there’s no support for most video filters (unless you’re using `auto-copy`)
Do you know if/why on apple silicon Macs there’s a difference between videotoolbox and videotoolbox-copy ? I thought the SoC memory is shared and that there wouldn’t be a need for copying data from VRAM to RAM.
If you autodetect hw accel, even the slowest underpowered gpu might get returned, which could result in worstened user experience compared to just using cpu.
Not sure why but I've had lots of issues with worse performance (lagging/stuttering) on Fedora (I think kernel 6.10) with a Vega 64. Every year or so I'll give it a try and it just won't work well and I give up again.
Yeah, I just pacman-installed mpv and tried running it on a 4k video. My CPU fan immediately kicked on, and top showed multiple cores getting involved. Kept scrolling this HN post to learn about hwdec=auto. Coming from mplayer and vlc my immediate question is, "What tradeoffs are the maintainers trying to make by not making this the default?"
https://interfacinglinux.com/2024/01/10/hardware-acceleratio...