Intel Discontinues High-Speed, Open-Source H.265/HEVC Encoder Project

Phoronix's Michael Larabel reports: As part of Intel's Scalable Video Technology (SVT) initiative they had been developing SVT-HEVC as a BSD-licensed high performance H.265/HEVC video encoder optimized for Xeon Scalable and Xeon D processors. But recently they've changed course and the project has been officially discontinued. [...] The SVT-AV1 project a while ago was already punted to the Alliance for Open Media (AOMedia) project and one of its lead maintainers having joined Meta from Intel two years ago. SVT-AV1 continues excelling great outside the borders of Intel but SVT-HEVC (and SVT-VP9) have remained Intel open-source projects but at least officially SVT-HEVC has ended.

SVT-HEVC hadn't seen a new release since 2021 and there are already several great open-source H.265 encoders out there like x265 and Kvazaar. But as of a few weeks ago, SVT-HEVC upstream is now discontinued. The GitHub repository was put into a read-only state [with a discontinuation notice]. Meanwhile SVT-VP9 doesn't have any discontinuation notice at this time. The SVT-VP9 GitHub repository remains under Intel's Open Visual Cloud account although it hasn't seen any new commits in four months and the last tagged release was back in 2020.

Laid off

[BeepBoopBeep]

They were probably laid off ... lol

Re:

[laughingskeptic]

Definitely -- no company lays off 19,000 software engineers and then keeps the teams working free stuff.

Re: Laid off

[saloomy]

It is the symptom, not the problem. The visionaries and innovators are gone at Intel. They can not innovate anymore. They are just recycling the same architecture and leveraging the advancements at ASMR to make that same architecture faster. No new instruction sets. No AI cores. No power-capped and high performance cores in the same chip. Stupid power consumption. No advanced SOCs. PCI Lane limitations. Poor multithreading performance degradations. Intel chips are circa 2009 printed on 2024 photolithography

Re:

[BeepBoopBeep]

INTC does in fact have fabless customers, and "trying" to grow the fab business. My buddy in the semi business that works with all the semiconductor companies explained why they are struggling: INTC has for its entire life been vertically integrated, their workflows and processes from design->chip is totally INTC specific, so working with INTC means working the INTC way, and not the easier way working with TSMC or Samsung.

Re: Laid off

[vux984]

" No power-capped and high performance cores in the same chip."

Intel has had performance cores (p-cores), and efficient (e-cores) for a couple iterations now. Is this not that?

Intel's also advertising NPUs (neural processing units) for its newest lineup...

https://www.pcmag.com/news/the... [pcmag.com]

Isn't that "AI cores"? It's hard to take your post seriously.

Re: Laid off

[saloomy]

Which of those were innovated by Intel? Itâ(TM)s just a copycat company. They innovated the 8086. The 386. The mmx instruction set. The SSE instruction set. All of those enabled new applications. None are invented now at intel.

Re:

[_merlin]

MMX was a poor man's VIS - they blatantly ripped it off from Sun. It's the same "overlay integer SIMD on FPU registers" crap, except they somehow managed to make it even worse than Sun's implementation. What were the innovative features in the 8086? The 20-bit address space was barely an upgrade over the 16-bit address spaces on the 6502 and Z80 families, and the segmented memory model ended up being a liability that made extending the address space any further far more difficult than it needed to be.

Re: Laid off

[saloomy]

The 8086 was the first CPU to have an x86 instruction set. If you consider that innovative.

Re:

[_merlin]

I don't think the 8086 architecture is innovative. I think Intel has had moments of brilliance, like LightPeak which became Thunderbolt. PCI was also a solid design, taking some of the ideas from NuBus and using reflected wave signalling for better stability and lower power consumption at higher frequencies. PCI Express is pretty good, too.

Even the 4004 was pretty innovative, particularly with the trade-offs it made to make circuit board designs simpler and scalable. The iAPX432 CPU was innovative, alth

Re:

[_merlin]

advancements at ASMR

Please enlighten me on this.

Re: Laid off

[saloomy]

Smaller pathways are enabled by using new fabrication methods all developed by ASMR. When you see chips go from 7nm to 5nm, that is ASMR figuring out new ways to print on chips.

Re:

[_merlin]

You should look up what ASMR is before continuing to make a fool of yourself.

Re:

[thegarbz]

No for-profit company develops free stuff period. They were doing this because they saw the ability to make a buck elsewhere.

Remember Indeo Video? Nobody else does either.

[Seven Spirals]

That's because it's a proprietary piece of horseshit and barely got used. Good luck with your private efforts, but here be dragons. You're in for a rough ride, Intel.

Re:

[Powercntrl]

Early PC software video codecs were all pretty bad. It basically just came down to a choice of what sort of terrible compression artifacts style happened to be your preference. I remember finally getting a Pentium (I'm referring to the socket 7 kind, well before Intel resurrected it as a brand) which could finally decode MPEG1 in software, and that's the first time video actually looked like video.

Re:

[nightflameauto]

Early PC software video codecs were all pretty bad. It basically just came down to a choice of what sort of terrible compression artifacts style happened to be your preference. I remember finally getting a Pentium (I'm referring to the socket 7 kind, well before Intel resurrected it as a brand) which could finally decode MPEG1 in software, and that's the first time video actually looked like video.

And the only reason I had any idea Wheezer was a band was my first Pentium tower, from my employer at the time, Gateway 2000. What a thrill it was to see a music video on a computer! Even if it was a terrible song, that I now have forever embedded in my brain because of all the techs playing it over and over and over again as a "troubleshooting" step for everything from bad CD ROM drives to video cards to audio cards to monitors. I swear that pack-along video on the Windows disc made that band.

Re:

[Seven Spirals]

Oh I hated those guys and cursed M$ even harder for including them because, as you say, that crap got played for all kinda reasons. I just looked though and that old video is CINEPACK not Indeo, but close enough.

Re:

[TheRealMindChild]

MPEG decoder cards were cheap and plentiful at that point. I remember purchasing an ISA MPEG decoder card at a computer show for $5 for my 386 machine

Proprietary even if open source

[BlueCoder]

The money drives it. It either gets incorporated in other projects or someone with time, motivation, and personal interest takes the reigns that otherwise would have been squeezed out by the corporate mentality.

It did what it needed for the time which was to sell server processors for cloud encoding.

AV1

[markdavis]

H.265/HEVC is being kicked to the curb by everyone at this point. The licensing, pricing, encumberments, and risk are all just too high. MPEGLA got too greedy and shot themselves in the foot. Lots of big companies have jumped ship, and together, designed their own, completely open and royalty-free specification, AV1:

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

The problem now is just waiting for hardware encoders/decoders to catch up. Most all video chips/cards now have AV1 hardware decoding support, but it will take a while for those with older systems to eventually upgrade. Most also have encoding support as well. From what I understand, AV1 is equal to or better than HEVC in every way, except encoding speed, and that has been rapidly improving. It took a long time to develop and get it going, but it is now starting its trajectory to be "the" standard for general video storage/streaming/etc.

Re:AV1

[tlhIngan]

.H.265/HEVC is being kicked to the curb by everyone at this point. The licensing, pricing, encumberments, and risk are all just too high. MPEGLA got too greedy and shot themselves in the foot.

Except it's not MPEG-LA. In fact, MPEG-LA wasn't greedy enough, so a bunch of patent holders went and formed the HEVC Alliance because they hated how little the MEPG-LA wanted for their patents.

So it's the fact licensing requires two licenses and one is far more expensive than the one you had.

AV1 is not universal yet - it's only been in hardware this current gen of SoCs, which means it's in maybe the flagship SoCs, but not in the midrange and lower. So those users will likely not have hardware AV1 for another 3-5 years at a minimum.

Re:

[loonycyborg]

Software decoding good enough anyway, unless you need insane resolution.

Re:AV1

[markdavis]

>"Except it's not MPEG-LA. In fact, MPEG-LA wasn't greedy enough, so a bunch of patent holders went and formed the HEVC Alliance because they hated how little the MEPG-LA wanted for their patents. So it's the fact licensing requires two licenses and one is far more expensive than the one you had."

Thanks for filling in the details. I didn't know exactly which organization(s) screwed the pooch.

>"AV1 is not universal yet"

Agreed. I only projected it is certainly on the path to be universal.

>" it's only been in hardware this current gen of SoCs, which means it's in maybe the flagship SoCs, but not in the midrange and lower. So those users will likely not have hardware AV1 for another 3-5 years at a minimum."

I am not up on current AV1 hardware compatibility, but, the ThinkPad X13 AMD Gen 3 laptop (AMD Ryzen 5 Pro 6650U, Radeon 660M) I bought almost 2 years ago (which had been out a year, I think) has AV1 hardware decoding.

On my AMD Ryzen 7 3700X Linux desktop, which is older and has an ancient Nvidia card it in, there is no hardware AV1 decoding, but the CPU is fast enough that Firefox software-decodes AV1 YouTube videos with no problems at all (about 20% CPU on one core for 720P, 30% for 1080P). Maybe someday I will get a new video card (maybe an AMD one).

Now, ENCODING AV1 would be painful, indeed, without hardware acceleration :) I am still encoding my own stuff to H.264.

Re: AV1

[Ken_g6]

SVT-AV1 on Handbrake has gotten much, much better lately. I switched to encoding with it and it can encode much smaller in the same time, or half the size of H.264 in three times the encoding time.

Re:

[thegarbz]

it's only been in hardware this current gen of SoCs

That is false, I mean not very false, but partially it does go back a couple of generations on general purpose stuff and longer than that on special purpose stuff. E.g. Qualcomm introduced AV1 decode in 2022. Samsung SoCs 2021. Intel in their CPUs in 2021. NVIDIA in 2021. Samsung and LG in their TVs in 2020. AMD in all GPUs and iGPUs in 2020. Mediatek in their SoCs 2020.

SO at this point if you bought something in the last couple of years chances are it has native AV1 decode support. Unless you have a Qualco

Re:

[drinkypoo]

H.265/HEVC is being kicked to the curb by everyone at this point.

All of the modern Nvidia cards do H.264, H.265, HEVC, and AV1. Usually when I want to do anything but just rip to the original format, It's because I'm trying to put content on a specific limited device. Then I have to use whatever codec it supports...

Re: AV1

[drinkypoo]

Oh look, I triggered my dick riding troll again. Good to know if I ever lose my home at least I can live rent free in their head.

Re: AV1

[Malc]

This just isnâ(TM)t true though. HEVC is doing very well and its usage is growing. Licensing hasnâ(TM)t turned out to be so scary and a lot of companies have decided to adopt HEVC. Even Google caved and enabled it in Chrome a could have years ago. Only in the niche open source world are people still pushing this misconception about HEVC.

SVT-HEVC was a poor implementation, even worse than x265, and didnâ(TM)t really stand a chance with everybody choosing Nvidia for their GPU-based encodin

Re: AV1

[jonwil]

If you don't need compatibility with existing things that use HEVC and aren't targeting hardware that can't do AV1 (or can't do AV1 and still get acceptable battery life) why would you pick HEVC with all the royalties and licensing?

Re:

[Malc]

While licensing is a concern, it's not the biggest expense for most professional content creators and distributors. Of course, we'd all like it if we didn't have to deal with this. Besides, AV1 isn't without its own patent and licensing issues - see for example Sisvel and the Advanci Video Pool. For private usage, nobody cares. As another example, anybody who's dealt with Dolby knows how horrible they are, but since their Blu-ray Disc loss to DTS, they've managed to weasel their way in everywhere. I wo

Re:

[thegarbz]

HEVC is doing very well and its usage is growing

Citation needed. We can find plenty of citations showing all major platforms accounting for the majority of video bandwidth are abandoning HEVC. Netflix, Disney Amazon etc delivery AV1 by default with a HEVC only as a fallback stream.

Even Google caved and enabled it in Chrome a could have years ago.

Enabling playback does not indicate support. Google infamously not only never used HEVC on Youtube, but was an early adopter of AV1 and just made AV1 the default playback format on Android. Their shift is clearly away from HEVC without ever supporting it internally.

Only in the niche open source world are people still pushing this misconception about HEVC.

Huh? I don't

Re:

[Malc]

Citation needed? I suggest you start with Bitmovin's 7th annual developer report [bitmovin.com]: nearly half of the people survey are using HEVC in production compared with about 8% for AV1 and 7% VVC. Yes VVC is that popular already.

This article has some numbers of video codec usage across major streamers. AV1 isn't as widespread as you think:
https://streaminglearningcente... [streamingl...center.com]

That article also talks about hardware support in mobile devices and the trouble Google got in to when they tried to enable software AV1 decode