Guesses on when AV2?

[benwaggoner]

So, it's still in early stages, but I was wondering when folks here think AV2 would have a final spec, and when it would start appearing in HW.

I've heard some strikingly different estimates, and so am interested in what the consensus here might be.

So, Doom9, when do you think AV2 will be done and available?

[Blue_MiSfit]

Is the question specifically when the spec will be done, or when hardware decoders will start to appear?

[benwaggoner]

I'm asking about both, but the poll is just for when the spec will be final and published, as it would happen first.

[Blue_MiSfit]

Cool, voted! (2024, for reference). I base this on absolutely nothing but my gut

[hajj_3]

AV1 hardware decoders took ages to appear. VVC/h266 was ratified 10.5 months ago, no news of hardware decoders for that yet. I think AV2 won't be released until at least 2024, waiting for h264 patents to expire in 2024-2026 would be very helpful. Regarding patents unifiedpatents has filed lots of lawsuits about hevc/av1 patents and has lost all the lawsuits that have been concluded apparently. So it seems that companies will likely have to pay sisvel to licence patents if they want to use vp9 or av1. Thankfully sisvel's fee is low.

[Spyros]

I agree that they will probably wait until several significant H.264 patents expire. The latest ones do so in 2027-2028, however the vast majority will be available to use by 2025, 22 years since the first version, 20 years since the addition of High/High 10 profiles (IANAL though, so I may be slightly wrong).

2025 also seems probable in order to be ahead of MPEG which may launch a competitive codec around that time. HEVC was standardized in 2013 and VVC in 2020, so a H.267 release in 2027 is not impossible. If at that time there are available AV2 hardware decoders / optimized software decoders (à la dav1d) it could make the adoption of the patented alternative less likely.

[cogman]

I voted for 2025, because I remember that the initial AV1 standard seemed to take quite a while to beat out. Since then, even more companies have come to the table which always seems to drag out a standard.

Hardware implementation timeline will depend on 2 things.

1. Getting current AV1 hardware implementations
2. How different is AV2 from AV1.

My assumption is that AV1 will be more like AV2 than it is different from it which will mean that hardware rollout shouldn't take as long as AV1 rollout did (though, I would have expected AV1 rollout to be faster given how similar it is VP9.)

With that said, for "not much different" rollout could be in less than a year. For "OMG, this is a whole new codec!" 2 or 3 years.

If the aom group wants to improve this situation, they could do what ARM does and publish reference hardware decoders. They certainly have companies with experience in doing just that.

[benwaggoner]

Originally Posted by cogman

I voted for 2025, because I remember that the initial AV1 standard seemed to take quite a while to beat out. Since then, even more companies have come to the table which always seems to drag out a standard.

Hardware implementation timeline will depend on 2 things.

1. Getting current AV1 hardware implementations
2. How different is AV2 from AV1.

My assumption is that AV1 will be more like AV2 than it is different from it which will mean that hardware rollout shouldn't take as long as AV1 rollout did (though, I would have expected AV1 rollout to be faster given how similar it is VP9.)

With that said, for "not much different" rollout could be in less than a year. For "OMG, this is a whole new codec!" 2 or 3 years.

Historically, Big New Codecs get adopted when there's a plausible path to a 50% bitrate reduction over the prior generation. That's MPEG-2 -> H.264 -> HEVC -> VVC. Codecs that have failed to impact the general A/V hardware market have generally be ones offering smaller gains (MPEG-4 ASP, VC-1, VP7-9, AV1 on the bubble). While lots of those codecs did get implemented in decoders, they didn't get adopted by the high eyeball-hour digital video markets like linear TV. And thus didn't get high quality realtime and offline encoders with good quality/speed tradeoffs.

The challenge for AV2 is to be enough better than AV1 technically and than VVC technically+licensing without taking so long that it competes with "well, H.267 is almost here." AV1 would be pretty much ignored on a purely technical basis; its attention is due to the free speech/beer promise and major web browsers hard coding out passthrough to HW for competing codecs. Technically, the most interesting thing about AV1 is mandatory film grain synthesis. But SoCs are being built with those at a different layer, making it pretty straightforward for HEVC or VVC to include AV1's FGS metadata and utilizing it for playback on any AV1 capable HW.

The HW decoder people I talk to are generally much more enthusiastic about VVC, and were sort of hoping they wouldn't have to do AV1.

If the aom group wants to improve this situation, they could do what ARM does and publish reference hardware decoders. They certainly have companies with experience in doing just that.

AV1 definitely suffered from low participation by decoder hardware design experts. Having them involved early could really help AV2. The complexity of AV1 decoding is pretty painful for both SW and HW decoders. I'd expect SW VVC decoders to outperform AV1 within a few years.