Comments on: VP9 Codec: Adopt it in your WebRTC application?

By: Tsahi Levent-Levi

Tsahi Levent-Levi — Mon, 02 Nov 2020 17:21:56 +0000

Behnam – thanks for these insights.

They are quite aligned with what I am hearing about VP9.

By: Behnam Kakavand

Behnam Kakavand — Mon, 02 Nov 2020 15:51:43 +0000

Hi Tsahi, thanks for the great post. Regarding the VP9 and HEVC they are both designed with HD and higher resolutions in mind, like 4K. With lower sizes they don’t offer much benefit. Also the libvpx is actually a very wild codec when it comes to the realtime encoding and its output bitrate ranges are all over the place without much of the bitrate and quality improvements over the H.264. So it is not only the matter of extra CPU usage, because all of those extra motion estimation and compensation an all other shiny tools need much more time or perhaps multiple passes to be effective. I am suspecting Chrome uses a specific implementation of the libvpx other than the opensource library, and this is just speculation. As for the AV1 currently it looks like the decoding complexity alone is on par with some encoding scenarios from H.264. For now the Dav1d looks to be providing one of the best decoder implementation and still it is very heavy. So I don’t believe the use of AV1 in webrtc or any other realtime usecase be a thing for a very long time, if ever. The most compelling candidate could be the MPEG 5 LCEVC, as an enhancement layer on top of either H.264 (or perhaps a better implementation of the VP9).

By: Tsahi Levent-Levi

Tsahi Levent-Levi — Wed, 30 Sep 2020 12:09:29 +0000

In reply to Angel. Thanks for sharing!

By: Angel

Angel — Wed, 30 Sep 2020 12:05:29 +0000

We tried VP8 and VP9 using Intel HW vaapivp9enc and vaapivp8enc in Gstreamer/Linux.
Finally we ended with VP8, because we could not make the vaapivp9enc to obey the output bitrate. It only worked in cqp configuration where HD/FullHD bitrate surged to over 11 Mb/s despite the limit of 3. VP8 behaved differently. Probably vaapivp9enc implementation has bug.
Unfortunately nobody among ATI/Nvidia does care about WEBvideo / VP9 encoding support despite the availability of a free HW encoder design for VP9. …

By: Vitaly

Vitaly — Mon, 31 Aug 2020 04:51:20 +0000

VP9 is not used because of its performance issues (I’m talking about libvpx now). It’s good on paper or in non-realtime / 2-pass encoding scenarios, but with -deadline realtime / –rt it’s actually A LOT worse than libvpx VP8, x264, x265, you name it

By: Tsahi Levent-Levi

Tsahi Levent-Levi — Sun, 16 Aug 2020 15:48:50 +0000

In reply to Pablo. I have no clue.

By: Pablo

Pablo — Sun, 16 Aug 2020 15:19:11 +0000

Does the iOS 14 VP9 implementation include hardware acceleration?
Is VP9 decoding more or less efficient than VP8?

By: Tsahi Levent-Levi

Tsahi Levent-Levi — Tue, 07 Jul 2020 14:41:51 +0000

In reply to Lennie. I don't think AV2 will be a thing in the next 5 years. I am also not sure how ready this VVC thing really is.

By: Lennie

Lennie — Tue, 07 Jul 2020 14:28:54 +0000

While I do think AV2 is coming at some point, but judging by the timeline on your 2018 AV1-codec blog post I think it’s a bit early to expect something this fast, 2022 was on my mind.

By: Tsahi Levent-Levi

Tsahi Levent-Levi — Tue, 07 Jul 2020 12:05:09 +0000

In reply to Lennie. Just saw that one myself. Wait 10 minutes until a press release is issued from AOMedia on AV2 just to keep us all confused.