El jue, 8 ago 2024 a la(s) 4:19 a.m., Mihai Popescu (mih...@gmail.com) escribió: > > Some people are really disappointed with VAAPI stuff and companion > VDPAU that they say it is better to avoid them all together and stay > on software side. > Complaints say the implementation is fixed in hardware thus not > flexible, and the drivers are really buggy and not maintained, so > anything based on them is rather not qualitative or breaks badly. Some > are obsolete already, since codecs change so often. > > I'm not a codecs expert, just reading what others say about this when > dealing with various implementations. >
Hi! In Linux, my experience with VDPAU is horrible. VAAPI, on the other hand, has better support. That said, as long as you're running Intel and AMD (with certain exceptions), the rest of the VAAPI drivers are pretty bad with support. VAAPI isn't a panacea either, in general its ability to speed up decoding is very good, something that I think many will be able to take advantage of and appreciate in OpenBSD, because it would make our machines consume less energy and give better performance. But in encoding, VAAPI isn't much help, unless you're using Intel. On my AMD RX 580, using hwaccel to encode using VAAPI+ffmpeg overloads the VCE and you're unable to encode 1080p60fps stable, when the GPU is capable of it (this is a well-known bug for ffmpeg). Curiously, if you use VAAPI+gstreamer you get rid of that limitation (in OBS they even created a gstreamer plugin to avoid the ffmpeg problem). Anyway, hwaccel is quite problematic, I understand that. Maybe with the arrival of Vulkan Video and its acceleration capabilities that will change a lot. ffmpeg 6 (using Vulkan and libplacebo) is capable of accelerating video, but that is something that only ffmpeg and mpv have the advantage of at the moment. Chromium already has a request to integrate that capability, but at the moment there is no work in that direction. -- "Dios en su Cielo, todo bien en la Tierra" ***********************************************
