TR Forums

https://code.facebook.com/posts/2538520 ... -use-case/

So the gist is (at least in FB's test) ~30-50% better compression, but longer encode times. Still, should be worth it for the big boys (Netflix, Google, Hulu, etc).

No H265 comparison, oddly.

meerkt wrote:

No H265 comparison, oddly.

That's the point. They don't want to pay the license fee. AV1 is free and purportedly better. It will take time to show up in use, though, because there is no hardware to decode it in consumer devices.

They're comparing it to H264, which is non-free as well. And the current commercial benchmark is H265.

AFAIK AV1 is about even to slightly better than HEVC at lower bitrates, HEVC taking the lead at higher bitrates.

IMO the the problem mostly is AV1 is second to market, and HEVC already has a solid foothold in devices with dedicated decode/encode hardware. It's in GPUs, camcorders, set-top boxes, DVD/BR players... Unless AV1 gets dedicated decode hardware it will actually cost the consumer more to use it just because of power use from less efficient decode. Netflix and Google will probably use it, but that's savings for THEM, achieved with additional costs (decode inefficiency) passed on to the consumer.

The royalty-free argument is hard to sell too, because that cost is born by the device manufacturers and isn't really felt by the consumer.

jihadjoe wrote:

AFAIK AV1 is about even to slightly better than HEVC at lower bitrates, HEVC taking the lead at higher bitrates.

IMO the the problem mostly is AV1 is second to market, and HEVC already has a solid foothold in devices with dedicated decode/encode hardware. It's in GPUs, camcorders, set-top boxes, DVD/BR players... Unless AV1 gets dedicated decode hardware it will actually cost the consumer more to use it just because of power use from less efficient decode. Netflix and Google will probably use it, but that's savings for THEM, achieved with additional costs (decode inefficiency) passed on to the consumer.

The royalty-free argument is hard to sell too, because that cost is born by the device manufacturers and isn't really felt by the consumer.

I'd agree with you except for the fact that literally ALL the big streaming names are onboard with AV1.

Apple, Amazon, AMD, ARM, BBC, Broadcom, Cisco, Google, Intel, Microsoft, Mozilla, Netflix, Nvidia. Of course it's going to take some time, but AV1 is no doubt going to replace HEVC(and VP9) in the not too distant future in a lot of places. It's still pretty early. There'll be hardware decode in another year or so.

H.265 won't die, tough. The movie and tv industry, which produce the majority of the content, are backing the incumbent and profiting from it. We, the consumers, will still need devices that support it. The biggest winner in this is the streaming providers.

DragonDaddyBear wrote:

H.265 won't die, tough. The movie and tv industry, which produce the majority of the content, are backing the incumbent and profiting from it. We, the consumers, will still need devices that support it. The biggest winner in this is the streaming providers.

Don’t forget that Apple, Amazon, and Netflix are now part of the “movie and tv industry” and have some pretty significant weight.

Please don't take offence, I just thought this was funny. OK here goes. Thanks MOV and quicktime : /

I'm also curious (and in agreement) that HVEC/x.265's lead is too significant for AV1 to make grounds on, especially if performance is only marginally better. The AV1 guys have also said that AV2 is in the works for a few years down the road, I wonder if they'll aim to do AV2 sooner rather than later...

What lead? Among who? It really only matters which corporations support which codec. More than likely, as h.265 support is already pretty well saturated, we will see AV1 added for HW decoding. And content providers will use whatever they use. Home users will still choose what they want.

If the only downside to AV1 is longer encoding times I doubt Netflix, Amazon, et al really care too much about that, but a reduction in bandwidth could be a big deal.

Codecs for streaming video have seen considerable churn in the last few years. If that continues then I'd expect it to start looking like the web API churn with most of the major players clustering around the most widely used technology which pisses them off the least.

The h.265 licensing situation is nothing like that for h.264.

For h.264, there is a single patent cartel selling licenses (MPEG LA), their prices for most uses are well within many companies' ability to pay, and many uses no longer require royalties at all.

For h.265, you have to pay expensive royalties to three separate patent consortia (MPEG LA, HEVC Advance, Velos) and, if you want to actually CYA legally, arrange licenses with fifteen other individual companies which hold h.265 patents but have not joined any consortium. This kind of disastrous legal mess and costly licensing situation is not something people are willing to deal with given that there are other viable options for video.

The founder of MPEG has admitted that this represents a complete breakdown of the MPEG royalty based format way of operating.

h.265 has failed. It didn't see significant enough adoption before the emergence of the superior AV1 to make companies willing to keep dealing with its licensing morass. This is the royalty-free formats endgame.

BTW, right now AV1 takes a horrifically long time to encode, but that's because the software work has been focused on making the right decisions for the format while finalizing the specification, not on encode performance. The format was designed with performance in mind and input from companies that are designing hardware encode and decode units. As they work on optimizing, it will speed up by around two orders of magnitude.

edit: adding a link to a comment I made at lwn early this year that says more about the legal issues involved.

Redocbew wrote:

If the only downside to AV1 is longer encoding times I doubt Netflix, Amazon, et al really care too much about that, but a reduction in bandwidth could be a big deal.

Codecs for streaming video have seen considerable churn in the last few years. If that continues then I'd expect it to start looking like the web API churn with most of the major players clustering around the most widely used technology which pisses them off the least.

This is accurate IMO.

The FB "study" cited 30-50% reduction in bandwidth for the same quality, but with longer encoding times. Which of course Google/Netflix/Amazon probably don't care much about due to their massive hardware stacks, and when hardware acceleration comes along in 2019 for a lot of end user devices things will get even better. I mean just look at that list of companies that are backing AV1.

Apple, Amazon, AMD, ARM, BBC, Broadcom, Cisco, Google, Intel, Microsoft, Mozilla, Netflix, Nvidia.

To be honest, AV1 seems nice, but kind of screams "too late" for it to be significant at this point given HEVC (H.265)'s foothold these days. Even my sort of ancient hardware (relatively speaking) will do hybrid decode of H.265 so things aren't too heavily loaded while a video is playing.

Significantly worse encoding performance will probably do it no favours in home use, either. I still stick mostly to H.264 mostly because of my CPU, and that the 970 doesn't really like H.265 files created by itself when played back by itself. (Maybe that's why NVIDIA never exposed H.265 encoding in GFE IIRC?) Its advantage in reducing bandwidth use over the network is probably very good to have, though.

Speaking from experience with 1080p content only. 4K might be different, of course.

Either way, one day I probably should start considering moving onto a new video card, but for the time being, things work.

Some of the talk seems to focus on a binary decision; YouTube will stream X, or it'll stream with Y. I imagine the path of least resistance for all involved will be the streaming provider querying your browser what hardware accelerated codecs can be used, and then stream AV1 if supported, otherwise HEVC for 4K or h264/MPEG-4AVC for 1080p and below. As hardware support grows, they get their desired switch to AV1. Maybe in a decade, they finally flip the switch on HEVC, etc.

They'll have no choice outside of browsers; phones simply don't have the raw power to decode 4K content unaccelerated, and STB's like Roku's wouldn't even try. And some of that consumer hardware, specifically STB's and streaming devices, will have a life of many years.

jensend wrote:

For h.265, you have to pay expensive royalties to three separate patent consortia (MPEG LA, HEVC Advance, Velos) and, if you want to actually CYA legally, arrange licenses with fifteen other individual companies which hold h.265 patents but have not joined any consortium. This kind of disastrous legal mess and costly licensing situation is not something people are willing to deal with given that there are other viable options for video.

New business opportunity: a service to maintain h.265 licensing for individual content creators.

Of course with AV1 it's a little late for that now.

Just for the note my ancient q6600 using windows 10 and Firefox av1 test (search for it) can play the video at 1080p with 30-40 max cpu, but is unable to play H265 or vp9 without hardware aceleration. this could help the fast adoption of the codec with no need to wait for hardware circuits in the GPUs.

Even my crappy Snapdragon 435 phone decodes HEVC 8-bit without issue.

The added encode compute complexity of the HEVC tier codecs will be mostly handed by server side of the streaming sites anyway. If for 99% of end users who are using a land-based Internet connection I doubt the codec choice matters at all, especially since everyone can just fall back to H264/VP9.

jihadjoe wrote:

AFAIK AV1 is about even to slightly better than HEVC at lower bitrates, HEVC taking the lead at higher bitrates.

jensend wrote:

the emergence of the superior AV1

How much more "superior" are we talking about here, I assume these are referring to quality?

strangerguy wrote:

Even my crappy Snapdragon 435 phone decodes HEVC 8-bit without issue.

The added encode compute complexity of the HEVC tier codecs will be mostly handed by server side of the streaming sites anyway. If for 99% of end users who are using a land-based Internet connection I doubt the codec choice matters at all, especially since everyone can just fall back to H264/VP9.

your phone video is hw acelerated as any phone cpu

try to play vp9 in youtube a 5 year old laptop

Flying Fox wrote:

jihadjoe wrote:

AFAIK AV1 is about even to slightly better than HEVC at lower bitrates, HEVC taking the lead at higher bitrates.

jensend wrote:

the emergence of the superior AV1

How much more "superior" are we talking about here, I assume these are referring to quality?

Yup, my data source was an Elecard comparison between AV1 and HEVC done a few months back before I made that post.

It suggests AV1 has a little bit of an advantage between 5Mbps and 15Mbps, but at the high-end (50Mbps+) HEVC has a big advantage over AV1

I would note, those figures look to be AV1 from Jan 2017. No idea what it looks like now as I haven't seen much news, but two years is a long time.

Still missing Samsung and Qualcomm, and that excludes two very large mobile SoC makers.

Otherwise putting it to use will be fairly quick, particularly if they can use non-dedicated hardware for acceleration i.e. OpenCL.

DancinJack wrote:

I would note, those figures look to be AV1 from Jan 2017. No idea what it looks like now as I haven't seen much news, but two years is a long time.

Honestly I don't expect AV1 to improve on quality, and I believe the bitstream has been frozen since the start of the year. Rather, quality will probably go down as they figure out how to cut down on the incredibly long encode times, so HEVC will likely remain the codec of choice for archival-quality stuff like Blu-rays. If we zoom-in however on the low-bitrate area, the advantages of AV1 are more apparent. This is from streamingmedia's look at AV1 from August this year.


The "Football" clip is supposed to represent the most challenging scenario (lots of detail, lots of fast motion). A VMAF quality of 93 is where the clip becomes free from visible artifacts and is considered the target. AV1 achieves that at a lower bitrate than HEVC (something like 2.05Mbps vs 2.7MBps).

Of course, the tradeoff right now is the incredibly long encode time:

At a factor of 45,316 x real time, current implementations of AV1 will take about 7 years+8 and a half months to encode a 90 minute movie on the reviewer's computer! (a Xeon E3-1505M v2) Even if we speed that up 8x (going from 2 cores to 16), and then x2 (from 2.8GHz to 5GHz+), it's still gonna take a hella long time to encode anything with AV1.

I understand that AV1 encoders right now are "reference" level, which basically maximize quality over anything else. Production encoders will likely dump some of that quality to gain speed, after which I suspect AV1 will lose a good portion of its quality advantage over HEVC, the remaining benefit being lack of royalties.

Oh for sure. It doesn't really seem like AV1 is gonna be for anyone but the biggest of boys like YouTube and Netflix and Hulu (etc). That is, of course, unless dedicated hardware for AV1 is like incredibly efficient.