AV1 video codec finalised

[AshleyAshes]

1 hour ago, yian88 said:

30% More efficient i assume 30% less size on disk and bandwidth required for stream, but that means nothing if the decoding performance is non-existent like hevc 4k 10bit ~50mbps, your PC will burn down with choppy video if the GPU doesnt support  it(my 970gtx doesnt and the cpu cant handle it).

Well, 4K HEVC is for sure doable on a CPU but you're firmly in quadcore territory to do it.  I've had 'no problem' doing 24hz 4K HDR HEVC (UHD BD Remuxed video stream) on the CPU of an i5 4590 and 3770k but that's 24hz and that uses 25-75% of the CPU.  Any 60hz simply burries the needle and drops frames, though my hex core i7 4930K can do 60hz but that's a lot of CPU to put into decoding video.

This all said, MOST consumer video decoding these days is not being done on big, fat, quad core CPUs.  It's phones, and tablets, and smart TVs and smaller laptops and consoles.  These devices require efficient ASIC or GPU decoding to make it workable.  Even if a lot of those devices HAD the CPU horse power, that'd be a real gas guzzler and it'd eat battery life for breakfast.

[LAwLz]

37 minutes ago, AshleyAshes said:

Well, 4K HEVC is for sure doable on a CPU but you're firmly in quadcore territory to do it.  I've had 'no problem' doing 24hz 4K HDR HEVC (UHD BD Remuxed video stream) on the CPU of an i5 4590 and 3770k but that's 24hz and that uses 25-75% of the CPU.  Any 60hz simply burries the needle and drops frames, though my hex core i7 4930K can do 60hz but that's a lot of CPU to put into decoding video.

This all said, MOST consumer video decoding these days is not being done on big, fat, quad core CPUs.  It's phones, and tablets, and smart TVs and smaller laptops and consoles.  These devices require efficient ASIC or GPU decoding to make it workable.  Even if a lot of those devices HAD the CPU horse power, that'd be a real gas guzzler and it'd eat battery life for breakfast.

Right now I am playing the Sony Camping in nature demo file using only CPU decoding (stock 1700X).

HEVC, average bitrate of 79Mbps (116Mbps peak), 4K and 60 FPS.

43 delayed frames and 4 dropped in total from beginning to end (most likely happened when I opened the task manager).

 

25-35% CPU usage.

 

Is there really such a big difference between our CPUs? What video player do you use? I use mpv.

 

I agree that hardware decoding will be necessary for widespread adoption though. Especially mobile devices.

[Coaxialgamer]

12 hours ago, Curufinwe_wins said:

I would like to know how high the encoding/decoding cost for this new codec will be.

 

HEVC is something like 10x more demanding than H.264. It happens that cpu/asic capabilities have managed to keep up fine (and thus I am using x265 on everything I can atm), but I'm not very sold on another potential 2-10x computational cost leap for another 30% reduction in size.

it's not really that much more demanding . It's just that dedicated hardware h.264 decoders have been a thing for a long time now, whereas the first hardware implementations of h.265 didn't really appear before pascal ( yes i know the gtx 960 has one ). Until now the difference was just that instead of using dedicated hardware , the CPu had to do it through software

[The Benjamins]

6 minutes ago, LAwLz said:

Right now I am playing the Sony Camping in nature demo file using only CPU decoding (stock 1700X).

HEVC, average bitrate of 79Mbps (116Mbps peak), 4K and 60 FPS.

43 delayed frames and 4 dropped in total from beginning to end (most likely happened when I opened the task manager).

 

25-35% CPU usage.

 

Is there really such a big difference between our CPUs? What video player do you use? I use mpv.

 

I agree that hardware decoding will be necessary for widespread adoption though. Especially mobile devices.

yours being a 8c16t CPU means that that video on a 3700k would be about 40-70% CPU.

[AshleyAshes]

14 minutes ago, LAwLz said:

Is there really such a big difference between our CPUs?

Well, if we go by CPU Mark, the 3770K averages 9531 marks where the 1700X averages 14633, so I'm gonna have to go with 'Yes, there really is such a big difference between our CPUs'?

 

The 1700X's CPUMark is comparable to my 6c16t 4930K.

 

And, again, the 1700X is a prime example of 'big fat desktop CPU' so I'm not sure what overall point you are trying to make.

[mariushm]

THIS (and 10/12 bit and Rec.2020 and HDR) is what the AV1 codec is targeted for ... good luck running the video linked fluidly (may run on gtx 1070 and gtx1080 but I'm not sure)

[Zodiark1593]

28 minutes ago, LAwLz said:

Right now I am playing the Sony Camping in nature demo file using only CPU decoding (stock 1700X).

HEVC, average bitrate of 79Mbps (116Mbps peak), 4K and 60 FPS.

43 delayed frames and 4 dropped in total from beginning to end (most likely happened when I opened the task manager).

 

25-35% CPU usage.

 

Is there really such a big difference between our CPUs? What video player do you use? I use mpv.

 

I agree that hardware decoding will be necessary for widespread adoption though. Especially mobile devices.

I'l have to try it on my Intel Atom system at home. It can barely handle low bit rate HEVC in 1080P with software decode.

[BuckGup]

3 hours ago, Mihle said:

Not really. My post was more ment at how a codec is used by companies rather than how the codec itself works. So what you say there is exactly what I ment that I want the companies to use the codec for, but probably not what they are going to do.

 

Companies would most likely use the codec to have the same noticeable quality for less size.

 

While I want them to rather use it to get better noticeable quality out of the same size.

 

I also wish the bitrate of like YouTube and Netflix was just higher to get better noticeable quality.

I agree 100%. YouTube compared to Vimeo is not even close. UHD Blu-ray compared to Netflix is miles apart

[LAwLz]

1 hour ago, AshleyAshes said:

Well, if we go by CPU Mark, the 3770K averages 9531 marks where the 1700X averages 14633, so I'm gonna have to go with 'Yes, there really is such a big difference between our CPUs'?

 

The 1700X's CPUMark is comparable to my 6c16t 4930K.

 

And, again, the 1700X is a prime example of 'big fat desktop CPU' so I'm not sure what overall point you are trying to make.

What I am saying is that if your quad core is at 70% CPU usage decoding a 24 FPS 4K video, then it seems like your decoder seems inefficient.

25% CPU usage on my computer = 60 FPS

70% CPU usage on your computer = 24 FPS

 

On CPU Mark my processor is 54% faster than yours, but the FPS and CPU utilization does not line up with that at all. Of course there are far more factors at play and you can't just look at it in such simplistic terms, but to me it seems like a slow decoder.

 

1 hour ago, The Benjamins said:

yours being a 8c16t CPU means that that video on a 3700k would be about 40-70% CPU.

How did you calculate that? Please remember that I am doing 60 FPS with that 25% CPU usage. She is doing 24 FPS with 70% CPU usage.

Even if we boost her FPS by 54% (which is the difference between her CPU and mine in terms of performance), we still have a massive gap of 37 FPS at 70% usage, vs 60 FPS at 25% usage.

 

Even if her CPU was at a stable 25% CPU usage it would still not line up with the expected performance.

 

 

 

1 hour ago, mariushm said:

THIS (and 10/12 bit and Rec.2020 and HDR) is what the AV1 codec is targeted for ... good luck running the video linked fluidly (may run on gtx 1070 and gtx1080 but I'm not sure)

Gonna download that and test it out.

Edit: And I am back. Yeah... 80% CPU utilization and dropping frames like crazy with CPU decoding.

You're not really correct in saying that AV1 is targeting 8K though. That is certainly one use for it, but it seems like AV1 will be suitable for low resolution content as well. 720p for sure, but maybe lower as well.

[AshleyAshes]

12 minutes ago, LAwLz said:

What I am saying is that if your quad core is at 70% CPU usage decoding a 24 FPS 4K video, then it seems like your decoder seems inefficient.

25% CPU usage on my computer = 60 FPS

70% CPU usage on your computer = 24 FPS

 

On CPU Mark my processor is 54% faster than yours, but the FPS and CPU utilization does not line up with that at all. Of course there are far more factors at play and you can't just look at it in such simplistic terms, but to me it seems like a slow decoder.

 

How did you calculate that? Please remember that I am doing 60 FPS with that 25% CPU usage. She is doing 24 FPS with 70% CPU usage.

Even if we boost her FPS by 54% (which is the difference between her CPU and mine in terms of performance), we still have a massive gap of 37 FPS at 70% usage, vs 60 FPS at 25% usage.

 

Even if her CPU was at a stable 25% CPU usage it would still not line up with the expected performance.

 

 

 

Gonna download that and test it out.

You're biggest mistake here is that you are assuming the CPU demands of all footage is identical.  This is not true.  There are multiple profiles within HEVC that deterine it's limitations but also which features are enabled which effect their computational load.  Your can't assume that the CPU load between two DIFFERENT clips will be the same just because they use the same base codec, resolution and frame rate.  Even the bitrate of a clip can matter since that ultimately determines how much data has to be processed for each frame.  You're attempting to make a serious apples to orange comparison and then asking someone to explain why some are red while others have an acidic taste.

[Curufinwe_wins]

1 hour ago, Coaxialgamer said:

it's not really that much more demanding . It's just that dedicated hardware h.264 decoders have been a thing for a long time now, whereas the first hardware implementations of h.265 didn't really appear before pascal ( yes i know the gtx 960 has one ). Until now the difference was just that instead of using dedicated hardware , the CPu had to do it through software

uhhh... cpu encoding and decoding for both without fixed-function is about 10x more demanding for h.265 for similar quality results.

 

And it sounds like this is another 3-5x more than that. Oh well. Eventually fixed function for it will come out and it won't matter.

[LAwLz]

6 minutes ago, AshleyAshes said:

You're biggest mistake here is that you are assuming the CPU demands of all footage is identical. 

You might have missed this part:

16 minutes ago, LAwLz said:

Of course there are far more factors at play and you can't just look at it in such simplistic terms, but to me it seems like a slow decoder.

The reason why I asked what player you use (which you haven't answered) is because I wanted to test it out on my computer, so that it would be an apples to apple comparison.

I think your CPU usage sounds very high, and by testing it with your player on my computer (using the same file I tested my setup with) I could see if it was your CPU just being much weaker, your decoder being less efficient, or your file being much more demanding.

 

 

5 minutes ago, Curufinwe_wins said:

And it sounds like this is another 3-5x more than that. Oh well. Eventually fixed function for it will come out and it won't matter.

Right now it's more like x300 times slower than HEVC, but that's without any optimizations.

It will be quite some time before AV1 is usable. Maybe a year before encoders start being somewhat decent, and then another year before fixed function hardware is out. So sometime in 2020 is my guess.

[Curufinwe_wins]

Just now, LAwLz said:

You might have missed this part:

 

 

 

Right now it's more like x300 times slower than HEVC, but that's without any optimizations.

It will be quite some time before AV1 is usable. Maybe a year before encoders start being somewhat decent, and then another year before fixed function hardware is out. So sometime in 2020 is my guess.

300x slower than HEVC. holy F. HEVC software decoding is already like 25% cpu usage for a 4k 10bit video on my OC 5820k.

 

Ok. Well lets come back to this when it is usable heh.

[AshleyAshes]

1 minute ago, LAwLz said:

Right now it's more like x300 times slower than HEVC, but that's without any optimizations.

It will be quite some time before AV1 is usable. Maybe a year before encoders start being somewhat decent, and then another year before fixed function hardware is out. So sometime in 2020 is my guess.

If I had a dollar for every useless thing that exists that someone once said 'Don't worry, in a few years it'll be amazing and everyone will be using it,' I'd have a sweet HD-DVD collection, all backed up to my Zune, with glorious ATRAC3 audio.

[LAwLz]

19 minutes ago, Curufinwe_wins said:

300x slower than HEVC. holy F. HEVC software decoding is already like 25% cpu usage for a 4k 10bit video on my OC 5820k.

 

Ok. Well lets come back to this when it is usable heh.

OK I might have exaggerated a bit (or maybe I am just remembering old results).

Results from a user on doom9 (for a 10 second clip):

x265 at the veryslow preset = 78s encoding time

aomenc on the cpu-used 2 preset = 1256s encoding time

So the AV1 encoder is about x150 slower right now. Bitmovin will demo an encoder that's 6 times as fast in just a few weeks though, so that number will quickly drop.

 

 

15 minutes ago, AshleyAshes said:

If I had a dollar for every useless thing that exists that someone once said 'Don't worry, in a few years it'll be amazing and everyone will be using it,' I'd have a sweet HD-DVD collection, all backed up to my Zune, with glorious ATRAC3 audio.

Yeah, because all these companies who have committed to supporting AV1 are clearly wrong and it won't take off...

Spoiler

• Amazon
• Apple
• ARM
• Cisco
• Facebook
• Google
• IBM
• Intel
• Microsoft
• Mozilla/Xiph
• Netflix
• Nvidia
• Adobe
• AMD
• Amlogic
• Argon Design
• Ateme
• BBC
• Bitmovin
• Broadcom
• CableLabs
• Chips Media
• Hulu
• Ittiam
• NG Codec
• polycom
• Realtek
• Sigma
• Socionext
• Veri Silicon
• Vidyo
• VideoLAN
• Xilinx

 

1) Why do you say AV1 is useless?

2) Why do you not believe it will take off?

 

The entire industry is fed up with the bullshit regarding video codec patents. HEVC was the straw the broke the camels back. Even Microsoft and Apple had enough. That should tell you what a massive issue it is.

[AshleyAshes]

Just now, LAwLz said:

1) Why do you say AV1 is useless?

2) Why do you not believe it will take off?

It's useless because it has no effective software or hardware decoders and has zero adoption.

I don't think it WON'T take off, I think that I have enough perspective to know that MOST things don't take off and it's irrelevant unless it DOES take off.  I'm not about to proclaim that a technology is the next coming of Jesus since so many before it have done so and only a minority have seen wide scale adoption.

[LAwLz]

Just now, AshleyAshes said:

It's useless because it has no effective software or hardware decoders and has zero adoption.

The bitstream isn't even frozen yet... What do you expect, hardware and software developers to have traveled back in time with complete solutions?

Software is already showing very big improvements in terms of efficiency, and hardware is expected to be out in 12-18 months. We know this because the codec has been developed in collaboration with Nvidia, Intel, AMD, ARM and other hardware vendors.

 

2 minutes ago, AshleyAshes said:

I don't think it WON'T take off, I think that I have enough perspective to know that MOST things don't take off and it's irrelevant unless it DOES take off.  I'm not about to proclaim that a technology is the next coming of Jesus since so many before it have done so and only a minority have seen wide scale adoption.

The reason why most things don't take off is because they get next to no support from the major players like Microsoft and Apple. WebP is a great example of this. Fantastic file standard. Killed by a lack of support in Windows.

This won't be the case here though, because (with the exception of Apple which joined very recently) this format was being developed in collaboration with all the major players.

 

This is not some hobby project like BPG which will lead nowhere because a lack of support. This is all the major players spending hundreds of millions of dollars to create a single format everyone agrees on using.

Do you not understand what a massive difference it makes to have native support in Android, iOS, OSX and Windows, compared to requiring third party plugins?

 

Also, AV1 will be the primary contender for NETVC. There are no reasons for NOT expecting it to take off.

[mariushm]

Fixed function hardware decoders will be available and supported in video cards and chips for TVs , phones etc ... majority of people change their phones every 1-2 years, maybe 3 years ... set top boxes from cable providers are easy and cheap to replace existing ones... people will buy new gaming cards or processors with integrated video cards every few years.... every 4-5 years or so most people will have new hardware.

 

The decoder for this new format is just a bit more complex, it's not a massive step in decoding complexity.. the encoder does the hard work.  So yeah, it will be eventually adopted, especially as Google pushes for it and will eventually replace it's VP9 encoded videos with this format

 

[AshleyAshes]

4 minutes ago, LAwLz said:

The bitstream isn't even frozen yet... What do you expect, hardware and software developers to have traveled back in time with complete solutions?

Software is already showing very big improvements in terms of efficiency, and hardware is expected to be out in 12-18 months. We know this because the codec has been developed in collaboration with Nvidia, Intel, AMD, ARM and other hardware vendors.

 

The reason why most things don't take off is because they get next to no support from the major players like Microsoft and Apple. WebP is a great example of this. Fantastic file standard. Killed by a lack of support in Windows.

This won't be the case here though, because (with the exception of Apple which joined very recently) this format was being developed in collaboration with all the major players.

 

This is not some hobby project like BPG which will lead nowhere because a lack of support. This is all the major players spending hundreds of millions of dollars to create a single format everyone agrees on using.

Do you not understand what a massive difference it makes to have native support in Android, iOS, OSX and Windows, compared to requiring third party plugins?

 

Also, AV1 will be the primary contender for NETVC. There are no reasons for NOT expecting it to take off.

Sorry, all I hear is someone proclaiming that an unproven and unadopted technology will TOTALLY be the second coming of Christ.

[LAwLz]

10 minutes ago, AshleyAshes said:

Sorry, all I hear is someone proclaiming that an unproven and unadopted technology will TOTALLY be the second coming of Christ.

Unproven - It has already proven itself to have significant bandwidth savings, and none of the licensing bullshit HEVC has. It has already proven itself to me as well as all the major software and hardware vendors.

Unadopted - It's not even out yet... Of course it's not widely adopted yet.

 

Also, stop with the strawman arguments. There is absolutely no argument for why anyone would expect this to not take off.

Whenever a format like WebP comes out the issue is always support. More specifically, support in Windows and third party software like the Adobe suit. That won't be a problem with this.

So why do you think this format is irrelevant? Do you have any reason for doubting the success of it? If so, please explain them to me. Saying that you have seen other formats fail is not a valid reason because those formats have not had the entire industry backing them up.

 

Edit:

Let me ask you this, do you think this will take off or not?

I am not asking you to see into the future and make an absolute statement. So please don't answer that you hold (what you believe to be) a rational position where you won't make any comments on anything until it is proven one way or the other. I am just asking what you think will happen. Do you think this codec will be widely adopted or not, and if you do how long do you think it will take?

 

I think it will, and I believe we will start seeing decent software and hardware for it in 2020. Then it will take people another 2-3 years to replace their old stuff for things that support it and start collecting a decent amount of media using it.

So by ~2023 I think it will be common to see AV1. Maybe about as common as VP9 and HEVC is today. That's my guess.

[Daegun]

23 minutes ago, AshleyAshes said:

It's useless because it has no effective software or hardware decoders and has zero adoption.

I don't think it WON'T take off, I think that I have enough perspective to know that MOST things don't take off and it's irrelevant unless it DOES take off.  I'm not about to proclaim that a technology is the next coming of Jesus since so many before it have done so and only a minority have seen wide scale adoption.

You should really think about the first sentence of this again. The standard that was finalized only days ago is useless because it isn't yet used in anything? That would be like saying that a specific gpu is useless before it was put in a card. You need the gpu first to make the card and likewise with this.

[AshleyAshes]

31 minutes ago, Daegun said:

You should really think about the first sentence of this again. The standard that was finalized only days ago is useless because it isn't yet used in anything? That would be like saying that a specific gpu is useless before it was put in a card. You need the gpu first to make the card and likewise with this.

This is ALSO true of every standard that FAILED to have mass adoption.  The argument is a fallacy.  It's useless until it see's mass adoption.  Not the PROMISE but -actual- adoption.  Plenty of standards have been backed by plenty of names and flopped.

 

And again, I'm not saying it for sure will NOT, just that it is silly to say it for sure WILL.

[LAwLz]

31 minutes ago, AshleyAshes said:

This is ALSO true of every standard that FAILED to have mass adoption.  The argument is a fallacy.  It's useless until it see's mass adoption.  Not the PROMISE but -actual- adoption.  Plenty of standards have been backed by plenty of names and flopped.

 

And again, I'm not saying it for sure will NOT, just that it is silly to say it for sure WILL.

You keep bringing up other standards which has failed in the past, as if they had anywhere near this level of support.

You think it was silly of me to question the efficiency of your decoder because it seemed like you used a lot of CPU resources? What you're doing is a far bigger apples to oranges comparison.

 

Also, I would like for you to answer my question I asked here:

1 hour ago, LAwLz said:

Let me ask you this, do you think this will take off or not?

I am not asking you to see into the future and make an absolute statement. So please don't answer that you hold (what you believe to be) a rational position where you won't make any comments on anything until it is proven one way or the other. I am just asking what you think will happen. Do you think this codec will be widely adopted or not, and if you do how long do you think it will take?

 

I think it will, and I believe we will start seeing decent software and hardware for it in 2020. Then it will take people another 2-3 years to replace their old stuff for things that support it and start collecting a decent amount of media using it.

So by ~2023 I think it will be common to see AV1. Maybe about as common as VP9 and HEVC is today. That's my guess.

 

Nobody here is saying that it will take off for sure. What I have said is that there is no reason to believe it won't take off. "Other formats have failed to take off" is an especially foolish argument because those other formats have not had anywhere near this massive amount of support.

 

 

21 hours ago, dalekphalm said:

Unless you can guarantee full industry adoption, including DRM support (sorry folks, it's a reality of things like Netflix)

Just noticed the DRM part.

Yes, it does support DRM. The same DRM that is currently used on other platforms.

[dalekphalm]

28 minutes ago, AshleyAshes said:

This is ALSO true of every standard that FAILED to have mass adoption.  The argument is a fallacy.  It's useless until it see's mass adoption.  Not the PROMISE but -actual- adoption.  Plenty of standards have been backed by plenty of names and flopped.

 

And again, I'm not saying it for sure will NOT, just that it is silly to say it for sure WILL.

Indeed.

 

Personally, if this does see mass adoption? Sweet! Kickass! Can't wait to use it!

 

Until then? Let's not overhype something that has yet to see any hardware adoption.

 

Plus, many people don't upgrade their CPU's or GPU's for a long time. Nor would they need to upgrade something like a streaming media box for a long time. Same with TV's. When I buy a TV, I want it to last 10+ years.

Pretty much the only reasonable argument is that mobile will see fairly quick adoption because people do tend to replace their phones somewhat quickly.

[Daegun]

32 minutes ago, AshleyAshes said:

This is ALSO true of every standard that FAILED to have mass adoption.  The argument is a fallacy.  It's useless until it see's mass adoption.  Not the PROMISE but -actual- adoption.  Plenty of standards have been backed by plenty of names and flopped.

 

And again, I'm not saying it for sure will NOT, just that it is silly to say it for sure WILL.

This is ALSO true of every standard that SUCCEEDED to have mass market adoption. See, I can emphasize something irrelevant as well. I wasn't saying that this will be adopted. I was saying that unused and useless are two different things and of course the standard that was just finalized is unused. 

 

And you have to see the humor in using a fallacy in the same paragraph you say I used one.