What's going on with EVC?

[benwaggoner]

My SMPTE colleague Michael Goldman just published a blog post about the recent MPEG codec standard releases and the general codec landscape.

There's obviously been lots of discussion here about AV1 and VVC, and some about LC-EVC (which is really a codec-agnostic postprocessing technology, not a codec).

But I've seen/heard very little about EVC here or other places. I've not deep dived on the testing results, but the claims are that Baseline is royalty free and better than H.264, and Main Profile is better than HEVC with much simpler licensing than either HEVC or VVC.

Anyone have any info or ruminations on EVC implementations, value, or potential market impact?

[videoh]

Originally Posted by benwaggoner

My SMPTE colleague Michael Goldman just published a blog post about the recent MPEG codec standard releases and the general codec landscape.

The link leads to a UPS site requiring an account. Is that what you intended?

I think most users are interested in formats present on physical disks and/or broadcast streams.

[ksec]

And for some strange reason the site doesn't work on Safari only on Chrome.

You may also want to use this link ( https://www.smpte.org/blog/a-new-compression-paradigm ) instead so you strip out any tracking information you had. ( Normally they dont cause any harm but we live in the age of privacy invasion, better safe than sorry )

[dapperdan]

They also mention that Main profile EVC is about the same efficiency as AV1 which seems to rather render the whole thing pointless.

Unless some absolutely fundamental part of both VP9 and AV1 is about to be taken out by an improbably successful patent attack then there seems little point in even considering it for any purpose, unless you have some ideological opposition to VP9 and AV1 that makes it worth taking the hit on quality, hardware suppprt and ease of implementation.

But good to see people embracing royalty free baselines as a sensible compromise after all these years, even if seems it's too late to save the MPEG codec ecosystem from itself.

[nevcairiel]

Originally Posted by dapperdan

They also mention that Main profile EVC is about the same efficiency as AV1 which seems to rather render the whole thing pointless

Thats why EVC has always seemed like some weird political/PR stunt instead of something that they expect to actually see wide adoption on. I guess future will tell.

[nhw_pulsar]

I am surprised that big players like Samsung and Apple for example doesn't seem to make the promotion of VVC and EVC.For now the ones who make the promotion/noise about VVC for example, are Fraunhofer and Nokia (also claiming a big part of its innovations, which can also be found in EVC Main).But maybe it is the tradition with MPEG codecs?, the big players will silently/quietly upgrade their products for sure with MPEG codecs, notably when hardware encoders/decoders will be ready?

[ksec]

Originally Posted by dapperdan

They also mention that Main profile EVC is about the same efficiency as AV1 which seems to rather render the whole thing pointless.

But good to see people embracing royalty free baselines as a sensible compromise after all these years, even if seems it's too late to save the MPEG codec ecosystem from itself.

EVC Main Profile should be much more energy efficient. Although by the time it arrives I am not sure it matters.

I am also extremely interested in Baseline profile, not a Royalty Free codec but truly ( on paper ) Patent Free Video Codec like Mpeg 2.

I am surprised that big players like Samsung and Apple for example doesn't seem to make the promotion of VVC and EVC.For now the ones who make the promotion/noise about VVC for example, are Fraunhofer and Nokia (also claiming a big part of its innovations, which can also be found in EVC Main).But maybe it is the tradition with MPEG codecs?, the big players will silently/quietly upgrade their products for sure with MPEG codecs, notably when hardware encoders/decoders will be ready?

I dont ever record Apple or Samsung make a big fuss about codec until they have hardware implementation. Which is still at least 2 years away.

[nhw_pulsar]

Originally Posted by ksec

I dont ever record Apple or Samsung make a big fuss about codec until they have hardware implementation. Which is still at least 2 years away.

Yes, I was just surprised at first glance that there is very little promotion of EVC and VVC compared to the big effort that makes AOM to promote AV1.But you're right, AV1 will be the Internet royalty-free codec, and EVC Main and VVC will be more hardware royalty-bearing codecs, and so then we certainly have to wait for their hardware implementation to start to hear some noise about them, which you estimate is in 2 years.

Let's come back on-topic, maybe also in the next 2 years we'll start to see some traction to choose EVC over VVC? Or is VVC definitely the MPEG-codec-of-choice?

[ksec]

Originally Posted by nhw_pulsar

Yes, I was just surprised at first glance that there is very little promotion of EVC and VVC compared to the big effort that makes AOM to promote AV1.But you're right, AV1 will be the Internet royalty-free codec, and EVC Main and VVC will be more hardware royalty-bearing codecs, and so then we certainly have to wait for their hardware implementation to start to hear some noise about them, which you estimate is in 2 years.

Let's come back on-topic, maybe also in the next 2 years we'll start to see some traction to choose EVC over VVC? Or is VVC definitely the MPEG-codec-of-choice?

Two years is really a hypothetical, optimistic, and unrealistic estimate. On the assumption that there are VVC "mobile" decoder hardware IP ( which there are none right now ). And Patent pool not being an issue.

As much as most of us enthusiast on Doom9 like to talk about Video Codec. It seems the market isn't that interested in a change. 80% of Video in 2020 is still in AVC / H.264 [1]. It might be the case that H.264 ends up being like MP3, it became good enough for the majority. Due to its complexity and quality ratio.
Which is why EVC is important, it promise something close to HEVC in quality without any complexity increase. ( Although There are no third party verification of that EVC results yet. )

Personally I just want the best codec to win, everyone pays a little more for it and we could just move forward.

But life is complicated.

[1] https://www.streamingmediablog.com/2...operators.html

[nhw_pulsar]

Originally Posted by ksec

Which is why EVC is important, it promise something close to HEVC in quality without any complexity increase. ( Although There are no third party verification of that EVC results yet. )

Very sorry if I misunderstood, but do you mean here that EVC Main will be a little better than HEVC in quality without any complexity increase? -You're not meaning that EVC Baseline is close to HEVC in quality but with the complexity of H.264?-

[ksec]

Originally Posted by nhw_pulsar

Very sorry if I misunderstood, but do you mean here that EVC Main will be a little better than HEVC in quality without any complexity increase? -You're not meaning that EVC Baseline is close to HEVC in quality but with the complexity of H.264?-

I mean the latter. EVC Baseline is close to HEVC but with complexity of H.264 and patent free.

[nhw_pulsar]

Originally Posted by ksec

I mean the latter. EVC Baseline is close to HEVC but with complexity of H.264 and patent free.

Ok! Indeed EVC is really very interesting!
-But I know it is very difficult from my humble little experience, because very quickly I have made a codec, for now there is just the intra (image compression) part but it is (visually) better than HEVC and very faster than H.264, but it never had any attention.-
Coming back to EVC topic, with these numbers, it could be game changing, but I have read for example that Google is involved to replace H.264 by VP9 on Internet, but would they change their plan for EVC? -This is an open question.-

[benwaggoner]

Originally Posted by ksec

And for some strange reason the site doesn't work on Safari only on Chrome.

You may also want to use this link ( https://www.smpte.org/blog/a-new-compression-paradigm ) instead so you strip out any tracking information you had. ( Normally they dont cause any harm but we live in the age of privacy invasion, better safe than sorry )

Good point, and so fixed.

[benwaggoner]

Originally Posted by dapperdan

They also mention that Main profile EVC is about the same efficiency as AV1 which seems to rather render the whole thing pointless.

I though the Main profile HEVC was suggested to be up to 40% better than HEVC. These early-stage comparisons always need to be taken with handfuls of salt, of course, as they are typically looking at mean PSNR or SSIM comparing fixed-QP, fixed GOP pattern encodes without any explicit psychovisual optimization.

[/QUOTE]Unless some absolutely fundamental part of both VP9 and AV1 is about to be taken out by an improbably successful patent attack then there seems little point in even considering it for any purpose, unless you have some ideological opposition to VP9 and AV1 that makes it worth taking the hit on quality, hardware suppprt and ease of implementation.[/QUOTE]
No philosophical objection to VP9 or AV1. VP9 was never a great fit for premium content due to the lack of encoders well-tuned for premium VOD scenarios as opposed to fragmented user-generated content with much lower quality expectations. AV1 already has a lot more third-party implementation and optimization than VP9 ever got. The bigger fundamental problem with AV1 is its designed isn't well tuned for efficient hardware decoder implementation, so the extra cost in SoC mm^2 is disproportionate. And despite lots of optimization, it's not great for efficient software decoding either. Also, squeezing out better encoding efficiency from AV1 requires a LOT more compute than MPEG codecs, in large part because they had to throw together a whole lot of patent-free tools instead of being to use the technically optimal tools like MPEG does. Of course, MPEG is what leads to patent licensing morasses like we've seen with HEVC.

The bigger problem with AV1 is that it doesn't really have any advantage over HEVC in terms of subjective quality @ perf @ bitrate. And I do have a philosophical/environmental objection to YouTube forcing lots of PCs to burn lots of extra watts using a SW decoder when HW ones are available. That's extra megawatts of electricity being used around the world, and the environmental impact thereof. That'll stop being a factor if/when AV1 HW decode becomes standard in real-world PCs. But we'd all be better off if YouTube would fall back to using HW HEVC decoders when an AV1 HW decoder isn't available.

Of course, encoding good AV1 takes a whole lot more compute as well, with its own environmental cost. For user-generated content with a much lower quality requirement, using ASIC encoders and turning off a lot of tools can be totally viable for that. As it is, everyone talking about 10% gains in AV1 over HEVC is also talking about >10x more encoding compute than HEVC in those comparisons. And HEVC can also look better if we can spend 10x more compute! That's why quality @ perf @ bitrate is the essential comparison point. Of course, the cost of perf goes down over time. And perf can get "wider" too - H.264 was able to take advantage of parallelization and SIMD more than MPEG-2, and HEVC can take advantage of parallelization and SIMD a lot more than H.264. On a >10 core AVX2 box, HEVC isn't that much slower than H.264, even though it is MUCH slower if just comparing single-core SSE2.

My hope for EVC is that, by threading the needle between the AV1 and VVC approaches, they've gotten a technically superior codec to AV1 but with acceptable licensing terms. Although we'd need a license and a lot of technical evaluation to determine if EVC is a great compromise, or unsatisfactory technically to companies fine with the MPEG approach and unsatisfactory to the "totally free or nothing" AV1 proponents.

But good to see people embracing royalty free baselines as a sensible compromise after all these years, even if seems it's too late to save the MPEG codec ecosystem from itself.

The success of the MPEG codec ecosystem is becoming more...attenuated. HEVC has been a runaway success in most of the industry, and hardware HEVC decode has been standard in new phones, TVs, streaming boxes, etcetera for years now.

Really the only place where HEVC HW decode hasn't become available by default is in Firefox and Chrome. And that's a political decision by Google and Mozilla to disable passthrough of HEVC to OS decoders, despite doing exactly that for H.264 (Edge and Safari do HEVC in-browser just fine). I get the political point they are making around patents, but I don't think the point needs to waste huge amount of electricity to be made. If AV1 does what it is supposed to do, it can succeed as a primarily hardware-decoded format, like H.264 and HEVC have been.

[Jamaika]

Originally Posted by benwaggoner

My hope for EVC is that, by threading the needle between the AV1 and VVC approaches, they've gotten a technically superior codec to AV1 but with acceptable licensing terms. Although we'd need a license and a lot of technical evaluation to determine if EVC is a great compromise, or unsatisfactory technically to companies fine with the MPEG approach and unsatisfactory to the "totally free or nothing" AV1 proponents.

I did a short test of vvenc v1.0.0 and xevc v1.0.0 codecs. The pictures show that the evc codec loses details compared to vvc.
VVEncoderApp.exe -i "113.yuv" -o "output_10bit.vvc" -v 5 -t 4 -s 1280x720 -r 29 -c yuv420 -g 16 -ip 16 -f 1000 --bitrate 3000000 --passes 1 --preset medium --level 6.3 --tier main
xeve_app.exe -i "113.yuv" -o "output_10bit.evc" -v 2 -d 8 -m 4 -b 15 -I 16 -s --input-csp 1 --codec_bit_depth 10 --output-depth 10 --rc-type 1 --bitrate 3Mbps -w 1280 -h 720 -z 29 -f 1000 --preset medium --level-idc 6.3
https://imgsli.com/NTY5OTA

[soresu]

[/QUOTE]The bigger problem with AV1 is that it doesn't really have any advantage over HEVC in terms of subjective quality @ perf @ bitrate. And I do have a philosophical/environmental objection to YouTube forcing lots of PCs to burn lots of extra watts using a SW decoder when HW ones are available. That's extra megawatts of electricity being used around the world, and the environmental impact thereof. That'll stop being a factor if/when AV1 HW decode becomes standard in real-world PCs. But we'd all be better off if YouTube would fall back to using HW HEVC decoders when an AV1 HW decoder isn't available.[/QUOTE]

Youtube isn't forcing anyone to decode AV1.

You can still disable AV1 streaming through your account settings precisely because HW ASIC decoding is not yet ubiquitous and therefore on many weaker phones and tablets using little cores would be unable to decode it well, let alone efficiently.

In which case I wouldn't be surprised if you get served VP9 or more likely AVC streams with it disabled.

Though actually with dav1d I've been told by a person working on it that they are already blowing past VP9 decoding efficiency on multi core systems due to much better provisioning for parallel encode/decode in the AV1 standard vs VP9.

It's not a case of "if" it will happen at all now for AV1 decode becoming standard for PC's, as all the major CPU/GPU ODMs Intel, AMD and nVidia have products supporting it in several market segments either in the market now, or less than a 3Q's to release on their roadmap (AMD Van Gogh and Rembrandt APUs).

Though even with AMD APU support currently lacking you can balance current ASICless APU's with either a RDNA2 or Ampere dGPU in platforms that offer them - so market availability willing (a large caveat currently) you should be able to build a brand new PC with AV1 decode support.

There's also at least one ARM SBC with AV1 decoding on the market now with the Amlogic S905X4, which will likely replace the S905X3 (or D3?) in the new Android based Chromecast soonish, with or without Google pomp.

As to us all being "better off if YouTube would fall back to using HW HEVC decoders when an AV1 HW decoder isn't available", even assuming that AVC wasn't viable due to bandwidth constraints that still assumes that Youtube also currently encode all videos in HEVC - do they?

I wouldn't be surprised if their pay to play videos for their Youtube/Google Play Movies and TV platform were also encoded in HEVC, but it seems pretty unlikely that the bulk of YT user generated content has any HEVC encodes at all these days.

I'm pretty sure the Argos encoding chip article also didn't mention HEVC at all, though I'll admit that doesn't necessarily mean that support for it isn't in there.

[benwaggoner]

Originally Posted by Jamaika

I did a short test of vvenc v1.0.0 and xevc v1.0.0 codecs. The pictures show that the evc codec loses details compared to vvc.
VVEncoderApp.exe -i "113.yuv" -o "output_10bit.vvc" -v 5 -t 4 -s 1280x720 -r 29 -c yuv420 -g 16 -ip 16 -f 1000 --bitrate 3000000 --passes 1 --preset medium --level 6.3 --tier main
xeve_app.exe -i "113.yuv" -o "output_10bit.evc" -v 2 -d 8 -m 4 -b 15 -I 16 -s --input-csp 1 --codec_bit_depth 10 --output-depth 10 --rc-type 1 --bitrate 3Mbps -w 1280 -h 720 -z 29 -f 1000 --preset medium --level-idc 6.3
https://imgsli.com/NTY5OTA

I don't think anyone expects EVC to even match VVC compression efficiency. "Better than HEVC licensing, better than AV1 performance, at least as good compression efficiency as either." seems to be the idea.

[Blue_MiSfit]

Pretty sure they don't encode any HEVC whatsoever. Why would they? Besides, that would erode their holy position

[benwaggoner]

Youtube isn't forcing anyone to decode AV1.

You can still disable AV1 streaming through your account settings precisely because HW ASIC decoding is not yet ubiquitous and therefore on many weaker phones and tablets using little cores would be unable to decode it well, let alone efficiently.[/QUOTE]
Disabling AV1 is something <1% of people would do. Something like 85% of people never even change one setting on their TV.

In which case I wouldn't be surprised if you get served VP9 or more likely AVC streams with it disabled.

That is exactly what you get.

Though actually with dav1d I've been told by a person working on it that they are already blowing past VP9 decoding efficiency on multi core systems due to much better provisioning for parallel encode/decode in the AV1 standard vs VP9.

Yeah, VP9 was really pathologically single-threaded. AV1 still isn't great, but much improved.

It's not a case of "if" it will happen at all now for AV1 decode becoming standard for PC's, as all the major CPU/GPU ODMs Intel, AMD and nVidia have products supporting it in several market segments either in the market now, or less than a 3Q's to release on their roadmap (AMD Van Gogh and Rembrandt APUs).

For PC, yes, I imagine we'll hit 50% AV1 HW decoders around 2025.

But PC is a relatively small slice of the global video consumption ecosystem, particularly for premium content. YouTube skews to PC way more than with paid content that can really benefit from a good TV and sound system.

The key lack in the broader market isn't gaming GPUs, but Qualcomm and Samsung. AV1 won't be a safe default codec to target for several years after Qualcomm adds AV1 decode to its lowest-end mobile chipset. We could well see VVC and AV1 support arriving around the same time in plenty of markets. And AV1 is pretty much ignored for markets and device categories where YouTube isn't that important.

Though even with AMD APU support currently lacking you can balance current ASICless APU's with either a RDNA2 or Ampere dGPU in platforms that offer them - so market availability willing (a large caveat currently) you should be able to build a brand new PC with AV1 decode support.

I think that's going to be quite slow deployment into 2022 due to fab shortages.

There's also at least one ARM SBC with AV1 decoding on the market now with the Amlogic S905X4, which will likely replace the S905X3 (or D3?) in the new Android based Chromecast soonish, with or without Google pomp.

As to us all being "better off if YouTube would fall back to using HW HEVC decoders when an AV1 HW decoder isn't available", even assuming that AVC wasn't viable due to bandwidth constraints that still assumes that Youtube also currently encode all videos in HEVC - do they?

They do not. Google begrudgingly added H.264 to Chrome ages ago, and hasn't added support for any new non-Google video codecs since then. But the incremental cost of encoding a decent looking HEVC is less than for a decent-looking AV1, so they could start doing so in steady state whenever.

I wouldn't be surprised if their pay to play videos for their Youtube/Google Play Movies and TV platform were also encoded in HEVC, but it seems pretty unlikely that the bulk of YT user generated content has any HEVC encodes at all these days.

You really can't play HDR in the living room without HEVC. Google has forced some companies to add VP9 decode in order to get YouTube certification, but I don't believe there has ever been a high volume VP9 or AV1 encoder tuned for excellent HDR quality. YouTube's economics don't overlap with excellent quality, but paid content absolutely does.

I'm pretty sure the Argos encoding chip article also didn't mention HEVC at all, though I'll admit that doesn't necessarily mean that support for it isn't in there.

I would be startled if anyone ever makes a SoC with a AV1 decoder but not a HEVC decoder. That's a chip that could never play a 4K Blu-ray, or ATSC 3.0 broadcast, or any streaming service that doesn't provide all their HDR content in AV1.

And encoding 4K HDR at a premium quality is >60:1 encoding time still. There's limited appetite for spending five days to encode a two-hour movie.

[ksec]

https://research.samsung.com/news/Op...l-Video-Coding