While it has been just over four years since its initial release, AV1 is maturing rapidly and beginning to see accelerated adoption, according to an expert panel, reports Mark Mayne.
The IBC webinar, titled ‘AV1 and streaming codecs’ saw representatives from Twitch/Amazon, Meta and Netflix come together to discuss the progress of the codec, the increasing adoption curve, and some of the key industry drivers that are feeding into ongoing development.
Ryan Lei, Software Engineer, Video Codec Specialist at Meta set out the stall on a positive note: “I think over the past few years, we have had hardware vendors start to include AV1 encoding support in their products. On the PC side we see Intel, AMD and Nvidia already announce hardware support, and also a lot of TV manufacturers have also added hardware decoder support.”
“On the mobile side we’re seen MediaTek and Samsung announce hardware decoding in their mobile SOC and recently Qualcomm also announced hardware support for their next generation Snapdragon SoC. So we’re hoping this adoption starts from the high end and then over the next, one or two years will trickle down to lower end mobile phones. As I said before, we have heavily optimised software that can already support encoding on most of the existing Android platforms. So I think ecosystem topology is definitely much better than a few years ago.”
Explosive growth in AV1 adoption
Krishna Rapaka, Video Researcher at Twitch/Amazon agreed: “The OTT workspace has been growing at a much faster rate for AV1, and it’s taking a lot of share from H265 HEVC. The numbers that I heard [anecdotally] are that there’s a 30% to 40% growth in AV1 adoption year over year, which is quite fast. But what we see is most of this is software driven at this point of time, but for services like Twitch hardware solutions are pretty critical because of the density and because of the scalability.
“However, what we are seeing is there is an uptick in hardware solutions recently and as more hardware solutions come up to the market, I think that will accelerate adoption pretty fast. We have seen some announcements from Nvidia, AMD, Xilinx and Qualcomm, which should help accelerate adoption.”
Ryan Lei agreed with Rapaka on the importance of hardware support: “The hardware support has been improving. When we started rolling this out, we used software decoders, which were instrumental to start streaming AV1 because you need to start streaming so that people could actually test AV1 on their devices and software decoders really helped a lot at this [early] stage. Currently we are seeing more and more penetration in hardware decoding, which I think is a good sign and we’re constantly working on working with vendors to certify [more] devices.”
The power of the ecosystem
AV1 was designed by the Alliance for Open Media (AOMedia) as a royalty free and open source successor to VP9 - that Alliance beginning in 2015 with seven founding members - Amazon, Cisco, Intel, Microsoft, Mozilla, and Netflix. These founders have since been joined by many other members including Google and Apple, and the AV1 codec is in active use by Netflix and YouTube among many others.
The panel agreed that the second key area that demonstrates AV1’s continued maturation cycle is ecosystem readiness.
Krishna Rapaka, Video Researcher at Twitch/Amazon said: “The cost of adding a new codec has a direct impact on the ecosystem. So if the ecosystem is completely ready, then it is worth adding the new cost of a new codec [and getting] the benefits of compression saving, whereas if the ecosystem is not ready, the benefit of adding a new codec is not scalable. What we see right now is that AV1 is a very new codec. Compared to H264, AV1 has made significant progress in the last few years on the ecosystem side. We are hoping that similar progress happens for products like H265 HEVC and H266 VCC, which is definitely something which we will try to evaluate. We are open to new products.”
AV1 uses “block-based frequency transformations” for encoding, based around VP9’s solution but with several enhanced techniques that result in a stream that needs less bitrate (and thus bandwidth) for a given image quality. That improved efficiency makes streaming HDR 4K video with wide colour gamuts more realistic, as well as potentially supporting 8K when needed. The downside is that achieving this efficiency is more computationally intensive than H.265 HEVC, increasing hardware overheads.
Mobile first not mobile-only
Andrey Norkin, Senior Research Scientist at Netflix was keen to rebut the impression that AV1 was specifically intended for mobile video: “I wouldn’t say that [AV1 is mostly intended for mobile video] - this impression is maybe because some companies rolled out AV1 first on Android. But the reason for this was really because of the decoders. You do need decoders and it was easier to support software decoding on mobile devices because of the device capabilities. When hardware decoding is supported on many devices I don’t see why it should be specifically targeted at mobile devices. It can work in other scenarios.”
Krishna Rapaka touched on the challenges of live streaming from Twitch’s perspective, and the crucial part that hardware support plays in the equation: “The traffic which we face is immense. So we have at any particular time 100-200k of channels and more than a couple of million concurrent users and to scale to this extent, what we need is hardware solutions, just to give a very high level overview. So once we ingest streams from streamers, we transcode them to multiple renditions, and to transcode them we do a lot of optimizations internally to make sure that the throughput is maintained, and that everything has been highly optimised. To achieve all of this, what we need is a hardware solution, which is so the software solution is typically a no go. So we [need to] look at a very high density, reliable hardware solution to solve this problem.”
No codec silver bullets
A final question asked whether the future held the possibility of a single, end-to-end encoding scheme, which Andrey Norkin responded to: “Well, it’s a pretty open question! I don’t know if there is just one scheme. In general there are certain codecs that support different parts of the pipeline. So for example, there are specific products like J2K, for example, that specifically target part of the pipeline, so probably it wouldn’t be just one codec that covers the whole ecosystem. AV1 you could say that it targets distribution rather than storage. So these are probably more prominent use cases. You can use it for image coding as well, but there are probably some parts of the pipeline where the targeted codecs are preferable.”