AV1 Beats x264 and Libvpx-Vp9 in Practical Use Case (facebook.com)

An anonymous reader shares a blog post by Facebook engineer: We tested AV1 (a new open-source, royalty-free media codec) under conditions that closely match the most common real-world use cases for Facebook video. Our test examined AV1's performance vs. practical open source video encoders that can be deployed to a practical production system, rather than merely testing efficiency vs. standard reference software encoders (i.e., H.264/AVC Joint Model or JM). By structuring the test this way, we were able to show how the codec will perform in a true production environment compared with current widely used alternatives, such as x264 and libvpx-vp9.

Our testing shows AV1 surpasses its stated goal of 30% better compression than VP9, and achieves gains of 50.3%, 46.2% and 34.0%, compared to x264 main profile, x264 high profile and libvpx-vp9, respectively. The new codec requires longer encoding times vs. current alternatives, however, due to increased complexity. Our tests were conducted primarily with Standard Definition (SD) and High Definition (HD) video files, because those are currently the most popular video formats on Facebook. But because AV1's performance increased as video resolution increased, we conclude the new compression codec will likely deliver even higher efficiency gains with UHD/4K and 8K content.

Comments

How about

[ArchieBunker]

AV1 compared to x265?

Re:How about

[Joce640k]

Also: What's the decoder efficiency? Does it have hardware decoding on mainstream CPUs/GPUs?

Re:How about

[afidel]

Royalty free and supported by all the major players goes a lot farther than a few percentage points so if it's even close AV1 wins. It also helps that they got the input of the hardware guys and that the algorithm is optimized for hardware implementations, combine that with the royalty free status and you can expect rapid uptake in the mobile space, something x265 has failed to achieve.

Re:How about

[crow]

I'm under the impression that AV1 is designed to be decoded using existing GPU shaders, so it's probably just a matter of software updates.

As to doing it on a CPU, I'm curious as to how it compares with x265, as I have a system that can't quite keep up with x265 in software for 1080p (older CPU and GPU). Would AV1 be better or worse?

Re:How about

[MBGMorden]

Time has shown that that for the most part "royalty free" doesn't mean a whole lot (ie, while Ogg and Matroska are perfectly functional, they never became the dominant forms used in their sectors).

x265 is catching on slowly because a lot of things don't have hardware decoding for it yet and it's pretty intensive to do using CPU alone. If AV1 isn't any better in that regard, then it won't gain any ground compared to x265 there.

Re:How about

[thegarbz]

May go further than a few but let the numbers speak. Hardware H265 encoders are common. The algorithm is not the reason that encoders aren't in the mobile space.

Re:How about

[Anonymous Coward]

grow a sack.

Re:How about

[trabby]

Ogg and Matroska are contrainers not codecs. AV1 is a codec.

Re:How about

[bluefoxlucid]

Spotify uses Vorbis, although we're all trying to get them to switch to Opus.

Similarly, I want the FDR Museum to switch its sound archives to Opus. I might buy their lossless original digital files, de-noise them with Audacity, Opus them, and republish them so the library can have the cleaned-up versions. Problem is they're $12.75 each and there are like 314 of them and they'd cost $4,000. Likewise, some of these aren't public domain and can't be rebroadcast.

Opus contains a CELP algorithm component that performs like Speex on speech data, and a general psychoacoustic algorithm like Vorbis on the remaining data. It's prone to vastly-superior compression rates at high quality. It's also better-suited for hardware decoding than Vorbis (uses a lot less RAM).

400 hours of speech would likely encode to little over a gigabyte at decent quality. 400 hours of music is more like 15 gigabytes.

Re: How about

[Type44Q]

They say "don't knock it 'till you try it..."

... sounds like you tried it.

Re:How about

[Kjella]

Time has shown that that for the most part "royalty free" doesn't mean a whole lot (ie, while Ogg and Matroska are perfectly functional, they never became the dominant forms used in their sectors).

Well Ogg and Matroska are container formats, if you mean Vorbis (2000) it was extremely late to the party compared to MP3 (1993) and missed the whole revolution. You can use Matroska with H.264/HEVC, if you mean Theora, the VPx codecs, Dirac and the other alternative codecs I'll agree they lost badly to H.264 and I think it's mostly because they never moved out of the software research stage and into accelerated hardware. After Google bought On2 then VP9 became something of a trial balloon for that through Android, now it looks like a broad alliance is ready to kick MPEG to the curb. I wouldn't bet on history repeating itself this time.

Re:How about

[Anonymous Coward]

Matroska is very much alive and popular today. It's used by YouTube for example. Granted, it's just a subset of it called webm, but webms are still Matroska files.

Re:How about

[Dwedit]

H.265 is only usable by Big Media Companies and Pirates. Media companies are vertically integrated and can deal with the horrible licensing fee situation, and Pirates simply don't care.

Re:How about

[TigerTime]

I don't think you understand the support behind AV1 vs x265. Only Apple is going to implement x265 in browser. The others, no. They ALL will implement AV1 in the browser. Netflix will be AV1. YouTube will be AV1. Facebook will be AV1. Amazon Video will be AV1.

The licensing around x265 is a complete disaster. Just about every major media company is going to steer clear away from them where possible. This is a completely different scenario than x264 vs the lesser known media formats.

Bottom line is the difference is support/backing. And it's vastly in favor of AV1.

Re:How about

[thegarbz]

Does it have hardware decoding on mainstream CPUs/GPUs?

How many CPUs and GPUs went through a complete design / test / ship to market phase in the past 9 days? AV1 literally just got it's specification out.

Re:How about

[thegarbz]

Ogg and Matroska are contrainers not codecs. AV1 is a codec.

Thanks Mr pedant. "ogg" is also the typically shortened form of saying "Ogg Vorbis" the latter part of which IS a codec, and the comment still stands.

Re:How about

[darkain]

And to add, this isn't conjecture either. Every major player you can name pretty much is a member of the alliance that is building AV1 (not just supporting, but the actual developers of the codec). Just right off the bat, you have nVidia, AMD, ARM, Intel (hardware), Cisco, Broadcom, Realtak (networking), Microsoft, Apple, Google, Mozilla, Adobe, VLC (software), Facebook, Netflix, Amazon, Hulu (content publishers). And this isn't even a complete list by far.

https://en.wikipedia.org/wiki/...

Re:How about

[thegarbz]

if you mean Vorbis (2000) it was extremely late to the party compared to MP3 (1993)

AV1 is late to the party compared to x264. Both produced similar gains over the other. Both had similar level of importance for bandwidth at their times (god I don't miss my internet speeds from back then)

Re:How about

[Anonymous Coward]

Vorbis is used in .webm video it seems. Opus may replace it. E.g. VP8 + Ogg Vorbis, or VP9 + Opus may be common.

Re:How about

[mentil]

Samsung is notably absent from their list of members, and given they're the #1 Android phone seller, that could matter. They also make their own Exynos SoCs for their own phones (used in their European models), which would have to support AV1.

Re:How about

[darkain]

Exynos is just a branded name of an ARM CPU. The thing with ARM that is different than Intel/AMD is that ARM doesn't actually manufacturer CPUs, instead they just create the architecture and reference designs that are highly modular. All it would take from Samsung is to click the checkbox to include the AV1 module when working with their next ARM (Exynos) CPU reference design.

There are other major "notable absences" from the list too, but are included by default without name simply because of their association with other companies. For instance, Twitch is a growing powerhouse in streaming media, however they're not listed. BUT, Twitch is owned by Amazon, and would most likely fall under that name.

Re:How about

[grub]

I hope he finds his virgin.

Re:How about

[AmiMoJo]

Huh... Of all the things I posted today, I really didn't expect this one to trigger someone.

Re:How about

[Kjella]

AV1 is late to the party compared to x264. Both produced similar gains over the other. Both had similar level of importance for bandwidth at their times (god I don't miss my internet speeds from back then)

There's a few billion people who still have bandwidth issues streaming a single 4K Netflix video. And then there's mobile that has both bandwidth limitations and traffic caps. There's absolutely still a market for a better video codec and the lack of royalty fees is just the icing on the cake. But yeah, there's plenty markets where H.264/x264 is "good enough" and already established.

Re:How about

[AbRASiON]

That's not entirely the point of it. The point is to be as good or better than pretty much any codec out there, but entirely royalty free.

This is a generally good thing. Wish the internet / companies / people did more of this stuff for the betterment of everyone.

Hopefully, it works out, without dumb issues holding it back, as is always often the case, in too-good-to-be-true scenarios with hardware or software.

Re:How about

[AbRASiON]

Everything I've read so far, indicates it's currently, insanely slow.
If you thought 265 encoding was bad, this is significantly worse, by like a factor of 10.

Hopefully this is fixed over time.

Re:How about

[Artemis3]

No it doesn't. Vorbis was surpassed by Opus, there is simply no better lossy audio compression at this point, so everything went Opus instead of Vorbis, and for lossless Flac is still preferred.

Also the parent is correct, you can't mix containers with codecs, Matroska (.mkv) is very widely used with all sort of proprietary codecs, such as h265 and aac. You can do that with ogg as well, but people tend to prefer mkv for a variety of reasons.

Re:How about

[thegarbz]

There's a few billion people who still have bandwidth issues streaming a single 4K Netflix video.

There were a few billion people who had bandwidth issues with high quality music back then too. My only point is, that this needs more than open and better to get ahead.

For the record they have this. An industry backed by a couple of major players that is slowly fed up with MPEG-LA's shit is what will push this through, but that's it. Royalty Free codecs are dime a dozend. VP9 also is royalty free and offers gains over H.264, also is pretty much a non-player outside of Youtube.

Re:How about

[thegarbz]

Also the parent is correct, you can't mix containers with codecs

That is pedantically true, but no one hears "ogg" during a discussion on open source codecs and thinks the person is talking about anything other than Vorbis. By the way I take issue with your:

there is simply no better lossy audio compression at this point

For two reasons: a) the best form of compression is one that is usable. Support for Opus is borderline non-existant, and FLAC is also relegated to special purpose areas. b) you're playing right into the OP's point. Both these codecs effectively failed to get any kind of reasonable traction.

Matroska (.mkv) is very widely used with all sort of proprietary codecs

Widly used sounds like the only thing you download is fan-subs of Japanese anime. MKV is like FLAC in adoption. It has carved a specific niche and is rare outside of it. Even H.265 is rarely seen in the mkv container. The mp4 container is far more common both in use and in software support. So again as the OP said: Royalty free means shit.

Re:How about

[Tough Love]

while Ogg and Matroska are perfectly functional, they never became the dominant forms used in their sectors

Overlooking your terminology abuse... Ogg Vorbis is the standard for game audio now.

Any patent troll?

[aglider]

I am looking forward...

Re:Any patent troll?

[afidel]

The legal teams of 40 tech giants have gone over the spec for a few years, the chances of there being an unknown patent is unlikely. If there's going to be a troll attack it's likely to be an unrelated overbroad patent that's unlikely to stand up to re-review.

Re:Any patent troll?

[JackieBrown]

It's a shame how you feel about jury duty. You'd rather have professional jurors? Or just binding arbitration?

Re:Any patent troll?

[afidel]

Not since TC Heartland, now they have to go after the company where they are incorporated, and that's Delaware in the majority of cases.

Re:Any patent troll?

[Pinky's Brain]

In patent cases, sure. It's a fundamentally broken system, not wasting the time of innocent people in maintaining that system would be an improvement.

It's not going to get worse for it, in fact it would almost certainly get better ... because juries of lay people just make patent cases more of a crapshoot (an intended effect by the people bringing the lawsuit) and while the value of patents might be negative, the value of gamblers is worse still.

Re:Any patent troll?

[Agripa]

The legal teams of 40 tech giants have gone over the spec for a few years, the chances of there being an unknown patent is unlikely. If there's going to be a troll attack it's likely to be an unrelated overbroad patent that's unlikely to stand up to re-review.

Even if there are no third parties with relevant patents, how is this different than JEDEC and RAMBUS?

I wonder if someone with an interest in H265/HEVC could take advantage of the situation by at least sewing uncertainty.

Re: Any patent troll?

[SpammersAreScum]

The video appears to be an April Fool's joke. The post is certainly off-topic. But have you seen it in a lot of other story discussions? If not, it's not spam.

Re:Practical Use

[jedidiah]

People that do this for a living can probably afford the relevant royalties for using the bog standard stuff. The rest of us likely have no reason to stop what we have already been doing.

For a long time I had enough trouble dealing with my current encoder setup (especially with Apple products). Now hardware has finally caught up to that. Tossing all of that aside doesn't really make much sense.

This stuff requires baked in specialty silicon to be useful in most cases. It can take a long time for the necessary stuff to become widely used to the point where you can just take it for granted.

Troll?

[higuita]

It is just me, or this thread is being attacked by one (or several) trolls ... i never saw soo many useless/insults/troll posts vs real posts in slashdot

what, patent trolls are crying because they will lose money with this format?!

Re:Troll?

[JackieBrown]

I'm waiting for the Trump for prison troll post.

All these threads are just filled with trash. We notice it more on articles that we are interested in.

There is still a lot of good comments and it helps pass the time

Donâ(TM)t

[backslashdot]

Basically I want a royalty-free format that everyone of importance (Firefox, Safari, Chrome, Edge) gets behind. Performance is a secondary concern.
Its still up in the air whether Apple will support this. Its true they joined as a founding member but they havent saif anything. They could just be spying for all we know.

Re:Donâ(TM)t

[swilver]

I think you'll find hardware support for decoding HD video is quite essential. Doing real time software decoding of full HD (let alone 4k video) is very intensive, and only viable on some of the most powerful CPU's.

Re:Dangerous to use open source

[DarkRookie]

It is no worse than Windows or MacOS. Better in most cases.

Not something to brag about

[Eravnrekaree]

If it didnt beat 264, that would be a problem and quite a shock, since 264 is old technology. So, thats not an accomplishment at all, beating a 10 year old format. They are really after beating 265, I would hope.

Re:Not something to brag about

[DigitAl56K]

Beating x264 (the encoder) is a very big accomplishment, because:
  - It is very well recognized as the leading H264 encoder
  - An incredible amount of time has been spent optimizing it
  - It has been done while avoiding patent encumbered, which takes many well established algorithms/techniques off the table

Claiming that it's "not an accomplishment at all" tells me that you're extremely ignorant of the level of work involved here.

Re:Dangerous to use open source

[Eravnrekaree]

Ever look at the problems with Flash, IE, Edge, Windows, etc etc? Open Source has a better record. I don't think any large C/C++ program has never had a bug. Propreitary software has a history of being much worse.

h265/hevc

[DrYak]

Current benchmarks (mostly synthetic tests) already show promising advantages in favour of AV-1 (the previous /. on AV-1's official announcement has links. Here 's yet another)
i.e.: per bits, it managed to pack more information than H265/HEVC

Now the psycho-visual optimization needs to be tuned a bit (the compressor need to learn better *which* of the information to pack or drop for a given amount of bits, but in general AV1 allows for more). And Netflix and Google should release more of the quality oriented tests (subjective tests from actual humans, and from AIs trained to have a somewhat similar response to human's visual system).
(As AV1 was just released, it's compressor isn't finely tuned yet and might wasting bit on packing information that an actual human viewer wouldn't give a shit)
(just like back when it was release x265 compressor didn't perform as visually pleasing as the older and better tested x264 compressor)

Over all that isn't much as a surprise.

H265/HEVC is an already released codec with a history.
AV1 is the new comer released now and supposed to be the next generation.

H265/HEVC isn't AV1's main competitor
AV1 supposed competitor is the next gen codec that will come out of MPEG (JVET), but that one isn't any close to release (but is expected to perform similarily good as AV1 compared to H265/HEVC)
Also the licensing shitstorm of JVET will also need to get solved once it is released, whereas the whole purpose that sparked AV1 was to make it royalty free.

Re:Practical Use

[guruevi]

The royalties to the end user are minimal if non-existent. The fact that it takes 5000x longer just means that you can't do live encoding on current hardware. That is a show stopper right there.

Re:Dangerous to use open source

[JackieBrown]

Why would you even respond to such an obvious troll?

Re:Practical Use

[TigerTime]

Right now? It's not practical. But they haven't optimized the encoders, and there isn't any silicon developed for it yet.

It's like trying to play some intense video game without a dedicated video card and using software to generate the scenes using un-optimized code. You're going to have a bad time.

Give it a couple years until the hardware is out there, and we'll see where we're at.

I keep seeing AVI not AV1...

[gosand]

They couldn't come up with a better name for it this codec? I know AVI is a container, while this is a codec, but AVI and AV1 just look so similar and could be nearly identical depending on the font. Might as well make it AllI1 (that's A el el EYE one)

Re:I keep seeing AVI not AV1...

[grub]

Me too. They should call it something like D1VX or Xv1d.

Re:Practical Use

[Gr8Apes]

For a long time I had enough trouble dealing with my current encoder setup (especially with Apple products).

Hmmm, I've been encoding video on Apple and standard hardware for many years. H.264 isn't a problem with any remotely modern higher level CPU, even if it does not have baked in hardware support. H.265 is considerably slower without hardware support and doesn't really offer enough of an advantage to be worth the extra encoding time given the cost of storage. Maybe I'll change my tune on 4K HDR when I get there.

Re:What's the point?

[segin]

H.265 is the gold standard for video. H.264 is a deprecated last-gen codec.