Intel Partners with Netflix to Deploy AV1 CODEC as its 4K Backbone

[btarunr]

At The National Association of Broadcasters Show (NAB Show) today, Intel and Netflix announced a new high-performance video codec that is available as open source and royalty-free to content creators, developers and service providers. Scalable Video Technology for AV1 (SVT-AV1) offers performance and scalability in video processing.

"The SVT-AV1 codec offers both high performance and efficiency. And compared to today's most popular codec (H.264 AVC), SVT-AV1 can help service providers save up to half their bandwidth, delivering leading-edge user experiences that can be quickly and cost-effectively delivered at a global scale. This codec makes it possible for services ranging from video on demand to live broadcast of 4Kp60/10-bit content on Intel Xeon Scalable processors, including the recently launched 2nd-Generation Intel Xeon Scalable processor," said Lynn Comp, Intel vice president of the Network Platforms Group and general manager of the Visual Cloud Division.



Modernization of video software codecs for increased efficiency will help deliver rich user experiences and reach global scale, accelerating time to market and lowering costs for developers and service providers. SVT-AV1 is a software-based scalable codec offering the best trade-offs among performance, latency and visual quality when working with visual cloud workloads. SVT-AV1 performance advantages are based on the SVT architecture, which is a cohesive and highly optimized codec architecture that already has delivered multiple generations of codecs, including SVT-HEVC, SVT-VP9 and SVT-AV1. The new SVT-AV1 codec is unique in that it allows encoders to scale their performance levels based on the quality and latency requirements of the target applications - ranging from highest quality video on demand (VOD) to livestreaming use cases. The high-quality encoding and decoding in SVT-AV1 will enable developers working with visual cloud workloads to get them to market faster. The codec is optimized for video encoding on Intel Xeon Scalable processors.

"The SVT-AV1 collaboration with Intel brings an alternative AV1 solution to the open-source community, enabling more rapid AV1 algorithm development and spurring innovation for next-generation video-compression technology," said David Ronca, director of Encoding Technologies, Netflix.

The SVT-AV1 codec is available under a permissive BSD+Patent license, which will make it easy to adopt and commercialize. Developers can access SVT-AV1 here.

View at TechPowerUp Main Site

 

[Fx]

While I welcome it, Comcast isn't going to like this news.

 

[ZhangirDuyseke]

Another reason to use Intel! AMD sux

 

[Shihab]

While I welcome it, Comcast isn't going to like this news.

ISPs will probably be on a fence here. Lower bitrates means lower consumption, granted, but it also means less cost to maintain high traffic...

The people who'd be mostly pissed are the HEVC folk.

 

[danbert2000]

ISPs will probably be on a fence here. Lower bitrates means lower consumption, granted, but it also means less cost to maintain high traffic...

The people who'd be mostly pissed are the HEVC folk.

It's hard to tell without some actual testing, but if they're targeting half of the size for similar quality compared to H.264, then this is going to be very similar to H.265 but without the crazy royalties. H.265 is going to be integrated into ATSC 3.0, so OTA programming in the US will push the requirement that H.265 decoders are in every TV set. In that context, this is only going to save Netflix money, not TV purchasers directly. And Netflix will have to maintain H.265 copies of all of their 4K programming because there are plenty of current TV sets with integrated Netflix that won't have AV1 support. So basically, Netflix will never be rid of H.265 but maybe they will be able to limit the growth of their royalty payments over time.

 

[B-Real]

Another reason to use Intel! AMD sux

Poor guy, go breath some air.

 

[notb]

It's hard to tell without some actual testing, but if they're targeting half of the size for similar quality compared to H.264, then this is going to be very similar to H.265 but without the crazy royalties. H.265 is going to be integrated into ATSC 3.0, so OTA programming in the US will push the requirement that H.265 decoders are in every TV set. In that context, this is only going to save Netflix money, not TV purchasers directly. And Netflix will have to maintain H.265 copies of all of their 4K programming because there are plenty of current TV sets with integrated Netflix that won't have AV1 support. So basically, Netflix will never be rid of H.265 but maybe they will be able to limit the growth of their royalty payments over time.

Netflix is on AWS. Storage on Amazon is quite cheap anyway ($0.021/GB/month). And to be honest: they don't have that much data to start with. And it grows really slowly.
In their case network traffic and CPUs will be the most expensive elements of the price Amazon asks (by far).
In the end millions of subscribers watch the same movie, not the other way round.

 

[danbert2000]

Netflix is on AWS. Storage on Amazon is quite cheap anyway ($0.021/GB/month). And to be honest: they don't have that much data to start with. And it grows really slowly.
In their case network traffic and CPUs will be the most expensive elements of the price Amazon asks (by far).
In the end millions of subscribers watch the same movie, not the other way round.

I am assuming that storing every single minute of all of their movies in multiple quality settings, resolutions, and formats will make a dent in their capital expenses. Also, you forgot that they have to encode all of this new content, too. So bandwidth is probably their biggest capital expense, but storage and compute for storing and creating the videos is probably quite a bit of money too. I wish I knew how HEVC licensing worked for Netflix. If it's per stream, this will definitely help over the long run. But they still doubled their storage needs, and that's going to hurt any company.

 

[notb]

I am assuming that storing every single minute of all of their movies in multiple quality settings, resolutions, and formats will make a dent in their capital expenses.

I don't think so. It's really not that much data. It's just movies/TV. It's limited. There's just so much professional content created. You can find different estimates online, for example this article:
https://gigaom.com/2013/09/18/build...ix-streams-114000-years-of-video-every-month/
says 100-150TB in 2013. Even if it went up 10 times (safe enough?), we're still talking about fairly small amounts - AFAIK somewhere in the range of what Youtube users upload weekly...

At the moment 1PB of AWS S3 storage costs just $23k/month.
Of course they use multiple AWS regions. Of course they use backups.
I'd still be surprised if the total figure was above 50PB, i.e. ~$1 mln / month. Netflix monthly revenue is ~$1300 mln.


But more importantly, and I've mentioned that, is the growth. The amount of data Netflix adds every day is limited. A few movies? Few dozen series episodes? That's nothing for such a giant.
There are some relatively small internet marketing companies (tracking user activity) that produce and process tens of TB daily.

Also, you forgot that they have to encode all of this new content, too. So bandwidth is probably their biggest capital expense, but storage and compute for storing and creating the videos is probably quite a bit of money too. I wish I knew how HEVC licensing worked for Netflix.

Netflix's shift to AWS was a hugely covered story, with many articles describing the process and their computing needs. If you look around you may find some answers.

 

[hat]

There's no reason not to use a more efficient codec. I'd like to think there's various copies of each movie or series stored somewhere encoded into different formats... when you're Netflix, it wouldn't be too hard to store multiple copies of even an entire TV series, and let the service auto negotiate the best codec available. Bandwidth is going to be the largest cost here with millions of subscribers streaming shit every day. If you can reduce that by say, 30% across the board simply by using a more efficient codec, I'd be willing to bet suddenly disk space seems cheap. It also helps when the codec in question is royalty free and you're not paying a fee to someone to use h.265...

 

[notb]

There's no reason not to use a more efficient codec. I'd like to think there's various copies of each movie or series stored somewhere encoded into different formats... when you're Netflix, it wouldn't be too hard to store multiple copies of even an entire TV series, and let the service auto negotiate the best codec available. Bandwidth is going to be the largest cost here with millions of subscribers streaming shit every day. If you can reduce that by say, 30% across the board simply by using a more efficient codec, I'd be willing to bet suddenly disk space seems cheap. It also helps when the codec in question is royalty free and you're not paying a fee to someone to use h.265...

We had a few discussions about game streaming lately and some people had concerns about bandwidth.
So to all these people that worry too much: if Netflix moves to AV1 and lowers usage by 20-30%, savings would cover game streaming for years.
And that's just Netflix. Other movie services and, most importantly, Youtube are also working on AV1 implementation.

 

[hat]

Aye, we use more bandwidth than ever before with the explosion of streaming services. But, I'm sure there are many like me who were faced with running over our data caps and now have to pay extra for additional internet. Adding game streaming on top of it is going to make it even harder to stay within such limits. Not to mention latency issues that exist for game streaming, but aren't a big deal for Netflix. I can still buffer a movie with a crappy high latency connection, but I can't buffer a game.

 

[notb]

Aye, we use more bandwidth than ever before with the explosion of streaming services. But, I'm sure there are many like me who were faced with running over our data caps and now have to pay extra for additional internet. Adding game streaming on top of it is going to make it even harder to stay within such limits. Not to mention latency issues that exist for game streaming, but aren't a big deal for Netflix. I can still buffer a movie with a crappy high latency connection, but I can't buffer a game.

Well, yeah. But many don't have limited cable connection (I mean: it's very specific to USA and quite rare in other developed countries).
So maybe game streaming will just develop faster outside US.
You're a very traditional nation anyway, very focused on "owning" stuff. I bet Americans will have the most moral issues with streaming (losing hardware).
The more liberal, tech-oriented societies, e.g. in Japan, Korea and some European countries, should be a lot more open to this phenomenon.

As for the latency... once game streaming becomes mainstream, I can totally see ISP offering "gaming-optimized connections" that minimize the issue of latency (just like we have connections made specially for people that watch a lot of movies from major streaming services)

 

[Aquinus]

It's hard to tell without some actual testing, but if they're targeting half of the size for similar quality compared to H.264, then this is going to be very similar to H.265 but without the crazy royalties. H.265 is going to be integrated into ATSC 3.0, so OTA programming in the US will push the requirement that H.265 decoders are in every TV set. In that context, this is only going to save Netflix money, not TV purchasers directly. And Netflix will have to maintain H.265 copies of all of their 4K programming because there are plenty of current TV sets with integrated Netflix that won't have AV1 support. So basically, Netflix will never be rid of H.265 but maybe they will be able to limit the growth of their royalty payments over time.

Cloud storage is dirt cheap. 100GB of persistent cloud storage at Google is 4 dollars a month which is peanuts. Streaming that content is what is going to cost the big bucks, not storing the content in the first place.

As for the latency... once game streaming becomes mainstream, I can totally see ISP offering "gaming-optimized connections" that minimize the issue of latency (just like we have connections made specially for people that watch a lot of movies from major streaming services)

The speed of light isn't getting any faster and neither is the speed at which electrons travel. Unless there is a data center near you, latency will be a problem. The reality is that if a data center isn't within 500km worth of network hops, you'll probably notice the latency. Even for round-trip latency, something as low as 5ms to the server will be perceived as at least a 10ms delay assuming input is processed instantaneously. Take satellite internet for example, latency is crap not because the method of transit is slow, it's because satellites are far away and distance = latency.