Slashdot Mirror

← Back to Stories (view on slashdot.org)

Moscow State University Releases 10th HEVC Video Codec Comparison (compression.ru)

Posted by timothy on from the squeezing-things-down dept.

An anonymous reader writes: The Graphics and Media Lab Video Group of Moscow State University has released its tenth video codecs comparison. This latest comparison focuses on HEVC codecs and includes some non-HEVC codecs such as x264 and VP9. The report concludes that Intel's MSS HEVC Software codec leads the pack in the "fast transcoding" use case whereas x265 takes the lead in the "ripping" use case. VP9 compares favorably to the HEVC codecs in the fixed quality and the speed versus quality test cases. See the PDF version of the report for more details.

51 comments

  1. Can't wait until they become popular by HalAtWork · · Score: 1

    Bandwidth savings using these codecs are remarkable. I can't wait until more hardware support for low power recording and decoding on most devices, and wider app support on the desktop. I've been wanting to transcode some of my media to fool around with quality and size settings but haven't seen much support in common apps yet.

    --
    Twinstiq, game news
  2. There is a cost with all that by QuantumReality · · Score: 5, Informative

    Encoding times. You can't squeeze more in less space without cost, cost here are cpu cycles. Encoding times of higher quality stuff which i did was about 5-6 times longer than comparable quality of x264. But it's size was around 40-50% smaller than x264 one. There is a cost not only in encoding, there is a cost in decoding too, it takes a lot more cpu cycles to decode x265. Either way HEVC is VERY GOOD codec.

    1. Re:There is a cost with all that by Anonymous Coward · · Score: 1

      HEVC doesn't work with low power devices. I can't show a 1080p video on a 5 year old laptop, and it runs pretty clunky on one I bought last year. So before you go mad encoding everything in it realize that it uses about 2-4x the cpu to show the same video. This basically can break them for anything not running a discreet gpu. Pardon me if I sit out the "public beta" period of the codec. :)

    2. Re:There is a cost with all that by sanf780 · · Score: 1

      HEVC is not supported by most HW. This situation is a similar one as when 1080p started to become popular less than ten years ago. Software like KMPlayer were the thing to have installed. Either you have supporting SW and HW or you are going to suffer a lot. I just hope that most HW video decoding blocks in computers can be of any use for HEVC. I know most appliances like SmartTVs will not.

    3. Re:There is a cost with all that by Crowd+Computing · · Score: 1
      Most desktop computers are powerful enough play back HEVC through software decoding. Many newer ARM-based SoCs already include HEVC hardware decoding support for 720p and 1080p video, the most common video quality found in the web and the torrent scene. An example is the Qualcomm Snapdragon 410:

      1080p HD video playback and capture with H.264 (AVC)
      720p playback with H.265 (HEVC)
      DASH is supported

      Most of the major Chinese semiconductor design companies already produce mid-range SoCs capable of decoding 4K HEVC.

    4. Re:There is a cost with all that by Anonymous Coward · · Score: 0

      We've seen that FLIF image format some days ago. It's amazing in that you can get a pic with x% quality by downloading just (k*x)% of the picture. It's optimal when one is just downloading a thumbnail (for instance). Obviously, for closer examination of the entire image, just the remainder must be downloaded. Pretty nifty IMHO.

      Could similar concepts be used for videos?

      Also, CPU is kinda cheap. In modern days, some distributions are even choosing to abandon old architectures (586 and earlier) and just accept 686 processors -- some even dumped all 32-bit support and went 100% 64-bit -- video is one of the few areas in which such foolishness makes sense.

      Network time is getting better all the time in urban areas. All over the world on the countryside, though, speed is still a major concern. Even worse is latency, which might be a non-issue in videos, but it's a major annoyance in video communication a la Skype (*).

      For that reason, optimizing for transmission beats optimizing for CPU. Particularly encoding is less of a problem, because ISPs usually have very heavy-load computers -- even conversions will be made on-the-fly, if not already.

      Decoding could impact older PCs, so that would be a secondary avenue for research, provided network transmission has the priority.

      (*) TM or C their respectibe owners, not me. BTW, what would be a free (as in GPL) format or application I could use here as example? Preferably one that woul work in the main 4 or 5 distributions (not just Fedora nor only in SuSE nor Ubuntu-only, I mean).

    5. Re:There is a cost with all that by slaker · · Score: 3, Informative

      The first time I played an MP3, my 486 sputtered and couldn't manage an unbroken audio stream. The first time I played a DVD, I needed a dedicated daughterboard to handle decoding.

      Right now, HEVC needs decent hardware and encoding takes a good long while. But it does play back fine on everything I have sitting around, going back to 3rd generation Core i CPUs, even with just Intel graphics. The i3 NUC in my living room doesn't have any problem with it at all. My STBs can't do it, but I can hand transcoding off to Plex and then they're fine as well. Given another year and everybody well catch or surpass Amazon's FireTV and have support for it as well. At that point, just like MP3s, MPEG2 and x.264, we'll be back to taking hardware support for granted.

      --
      -- I wanna decide who lives and who dies - Crow T. Robot, MST3K
    6. Re:There is a cost with all that by Gaygirlie · · Score: 1

      HEVC is not supported by most HW.

      HEVC is actually pretty well supported by mobile - devices and "smart-TVs" provided they are 2-3 years old or newer. Support on desktops and laptops is still lagging behind, but it's getting better fast there, too.

    7. Re:There is a cost with all that by Anonymous Coward · · Score: 0

      "but it's a major annoyance in video communication a la Skype"

      Skype is laggy because Microsoft now routes all Skype calls through their data centers in the US. I have always assumed this is to facilitate access for the NSA and their ilk.

      Of course you may have bought the rhetoric hook line and sinker that since you're not a terrorist you have nothing to fear. Well, the main advantage the NSA gives the US government and economy is industrial espionage. DO NOT conduct business meetings over Skype, you are simply asking for trouble...

    8. Re:There is a cost with all that by Anonymous Coward · · Score: 0

      HEVC Is a crap codec, simply because of all the licensing issues and costs.

    9. Re:There is a cost with all that by Anonymous Coward · · Score: 0

      Of course you may have bought the rhetoric hook line and sinker that the NSA gives a shit about your activities. And of course you have ignored that one of the documents released to the public states the NSA abandoned it's bulk collection efforts on internet traffic. The amount of resources to handle that amount of data was excessive and did not produce any tangible benefits and the program was abandoned.

    10. Re:There is a cost with all that by tdelaney · · Score: 1

      Whilst this is currently true, the situation is improving rapidly. I've been periodically testing the OpenELEC Kodi Jarvis alpha builds on my Raspberry Pi 2.

      The previous time I tested it (a month or so ago), 720p HEVC was just playable - ~100% CPU on both cores, but only dropping the occasional frame. The time before that, 720p HEVC was unwatchable. But with build #1016 (which includes FFMPEG 2.8.1) I was getting smooth playback and averaging around 60% CPU on both cores.

      HEVC will obviously never have the same hardware requirements that h264 does now, but there is a lot of work currently going into reducing the requirements.

      Of course, I'd much prefer that royalty-free codecs take the fore.

    11. Re:There is a cost with all that by Anonymous Coward · · Score: 0

      Could similar concepts be used for videos?

      In theory, yes. In practice? Problematic for streamed video but should be relatively easy for pre-packed video like that on Youtube that is transferred over a TCP-based protocol anyway.
      For streamed video you will have to chop it up in chunks and use a transport layer protocol where you can prioritize resending packets based on content.
      The protocols and APIs for something like that are trivial. To get to work in practice requires some tweaking.

      For streamed video you would rather want to have a format where packet loss leads to quality loss rather than frame loss so that you can just keep on streaming without having to focus so much on getting the right packets over.

    12. Re:There is a cost with all that by Hodr · · Score: 1

      It certainly does, just needs to be newer hardware. I have a $130 Braswell based NUC that I play x265 media on (1080p, no 4k) without issues. And it runs sub 15 watts for the total system.

  3. X264 still relevant by sanf780 · · Score: 1

    I might be reading the partial report wrong, but as far as I understood, x264 is not scoring low against HEVC. Either that or HEVC encoders are not mature enough. I get that the "real-time" encoder has very specific constraints, but how about the others?

    1. Re:X264 still relevant by Kjella · · Score: 3, Insightful

      I might be reading the partial report wrong, but as far as I understood, x264 is not scoring low against HEVC. Either that or HEVC encoders are not mature enough. I get that the "real-time" encoder has very specific constraints, but how about the others?

      From what I understand there's three areas where HEVC does very well:
      - Extremely low bitrates, because there's more blurring and less blocking that looks less bad.
      - 4K/UHD resolution because it supports larger block sizes that are more efficient at high resolution.
      - Better parallelism (WPP) for software decoding, if you have full hardware support it doesn't matter.

      For moderate resolutions like 1080p at moderate bit rates x264 is still performing very well and it's highly optimized. HEVC brings some new tricks that should improve compression further, but nothing really revolutionary like 20-30%. I mean there's better picture compression than JPG and better audio compression than MP3 but they're "close enough", at this point there's no big need to hurry unless you control the entire ecosystem like say Netflix or YouTube and can do HEVC where it's supported and downgrade to H.264 where it's not. For encoding to a broad number of unknown devices it's going to be H.264 for a long time to come. Looking at broadband speeds most people will either have so fast connections it doesn't matter or so slow connections it doesn't matter, there aren't many in the gap where the size difference is really significant. And I got a feeling 100GB for 4K BluRay is plenty, there are many movies now using only 20-40GB of the possible 50GB so I suspect we'll see H.264 used quite a bit there too. And the commercial terms for HEVC are worse, so there's very little compelling need to use it really.

      --
      Live today, because you never know what tomorrow brings
    2. Re:X264 still relevant by Anonymous Coward · · Score: 1

      > I might be reading the partial report wrong, but as far as I understood, x264 is not scoring low against HEVC.

      That was my take away. At very, very low bitrates HEVC produced superior results, but once you got to around 4mbps the difference between x264 and the best HEVC encoders was negligible. For me, ~4mbps is the bare minimum for 1080p anyway, so that makes HEVC useless to me as of now. That is kind of disappointing, actually.

      > Either that or HEVC encoders are not mature enough.

      I am confident that is a major factor. The x264 encoder has well over 10 years of active open-source development during which quality improved substantially. The HEVC encoders don't have anywhere near that amount of real-world usage and refinement yet.

    3. Re:X264 still relevant by Anonymous Coward · · Score: 0

      I just ran a test comparing h264 and h265 using a DVD source. With similar settings the h265 file is visibly better in quality, especially around edges and the file was 753MB whereas the h264 file was 1043MB. Both very similar to the DVD source.

      So that's squarely 20-30% improvement over h264.

      Decoding both files in software (VLC) used about 20% CPU on my old 3.06GHz Core 2 Duo (from 2008).

      Encoding the h264 file took about 2 hours, encoding the h265 file took 3.5 days. h265 is a fantastic technical achievement, even when compared with h264. The only thing that might hold it up is licensing.

    4. Re:X264 still relevant by jedidiah · · Score: 1

      ...except encoding takes DAYS rather than hours. This is a detail that you even mentioned yourself but chose not to acknowledge as meaningful.

      Also, declaring that the settings were "comparable" is a bit disengenuous. Even back in the bad old days when I used divx for better encode speed, an h264 did not take DAYS.

      --
      A Pirate and a Puritan look the same on a balance sheet.
    5. Re:X264 still relevant by Anonymous Coward · · Score: 0

      except encoding takes DAYS rather than hours.

      Yes, use VP9. The encoding doesn't take days and you can get better quality than x264. Version 1.4.0 of libvpx included a number of speed ups for VP9 encoding. They used version 1.3.0 in this comparison.

    6. Re:X264 still relevant by Anonymous Coward · · Score: 0

      Nice of you to respond to an AC. Comparable in that the H265 stream was similar but slightly better output quality.

      There's lots of situations that encoding time isn't really important and I'm working on a rig from 2008- seven years ago. I'm sure something a little more up to date or with some hardware optimizations would be much faster. But again who cares if encoding takes a long time if you only have to encode once. If you need a faster encode use different settings- like TFA.

      BTW, if there's one word you should have taken the time to spell right it's "disingenuous".

    7. Re: X264 still relevant by Anonymous Coward · · Score: 0

      4 microbits per second isn't such a high bitrate.

  4. My x265 Experience - Poor by Anonymous Coward · · Score: 1

    My experience with x265 and 4K video has been quite poor. Transcoding speeds are slow, like under 10fps on a i7-4790K CPU @ 4.00GHz.

    Decoding/playback is barely OK. VLC playback of 4K HEVC video crushes the processor and there are far too many hangs/glitches.

    The fact that x265 doesn't, and perhaps won't, have any sort of hardware acceleration support just makes it awful to do 4K work.

    1. Re:My x265 Experience - Poor by Gaygirlie · · Score: 2

      The fact that x265 doesn't, and perhaps won't, have any sort of hardware acceleration support just makes it awful to do 4K work.

      x265, just like x264, are software-encoders; of course they're going to be slow. If you want hardware-accelerated encoding you could, for example, get yourself a GTX 960 and compile ffmpeg with NVENC - support. I have a GTX 970 and I've played around with ffmpeg with NVENC quite a bit and if my memory serves correctly I get around ~90 FPS when encoding 1080p - source using NVENC. I haven't tried with 4K - video, I'll grant you that, so I have no idea how fast it can encode that. I could try it, I suppose, if you really wanted me to.

    2. Re:My x265 Experience - Poor by Anonymous Coward · · Score: 0

      My experience (watching Tiny4K porn downloaded via torrents) with 4K video on Win7 using VLC has been abysmal. The player just can't manage it (2013 i7m) and the video is unwatchable. I downloaded and tried "Media Player Classic" and the playback is silky smooth and lovely, which was a relief, since Tiny4K is awesome.

    3. Re:My x265 Experience - Poor by Gaygirlie · · Score: 1

      I think there is some bug with VLC wrt. HEVC. No matter if I'm using H/W-acceleration or not VLC just can't play HEVC-videos smoothly, but if I'm using some other player it's silky-smooth even at 60 FPS.

    4. Re:My x265 Experience - Poor by Anonymous Coward · · Score: 0

      I haven't tried with 4K - video, I'll grant you that, so I have no idea how fast it can encode that. I could try it, I suppose, if you really wanted me to.

      I'd really appreciate hearing those results.Theory holds that 4K, or more accurately 2160p should be around 22fps on your rig. But, that's just theory.

      Please test and let us know. Also, are you doing this on Linux or Windows?

    5. Re:My x265 Experience - Poor by Gaygirlie · · Score: 1

      I'd really appreciate hearing those results.Theory holds that 4K, or more accurately 2160p should be around 22fps on your rig. But, that's just theory.

      Please test and let us know. Also, are you doing this on Linux or Windows?

      I am doing this under Windows and I seem to be getting 28 FPS while doing NVENC_HEVC. I am using a VBR 2-pass preset, which is the slowest, but highest-quality combination possible on a Maxwell 2 NVIDIA GPU -- dropping to 1-pass or using CBR would speed it up some. 28 FPS 4K H/W-based HEVC-encoding isn't bad, IMHO. On my system (Xeon 1230V3) CPU-utilization is sitting at 14% and as I have SpeedStep enabled the system hasn't even bothered to raise the CPU's clocks for this, they're sitting at the lowest speed they can go -- basically, the system is perfectly responsive and useable while doing this encoding.

      I don't know what else you'd like to know, but I hope this was helpful.

  5. Re:Moscow State? by jonnythan · · Score: 0

    This will blow your mind:

    http://www.gram.edu/

  6. Still won't work in Internet Explorer by RogueWarrior65 · · Score: 1

    How the hell does Microsoft manage to get away so such dismal compatibility?

    1. Re:Still won't work in Internet Explorer by Anonymous Coward · · Score: 1

      No web browser supports H.265 (aka HEVC) video. In contrast Firefox, Chrome, Opera, and soon Microsoft Edge support VP9 video. So don't worry about H.265 for browsers. Just use VP9 and be happy.

  7. Patents by RenHoek · · Score: 2

    Doesn't matter how good x265 is, as long as patent litigation clouds keep forming over it, it will not succeed.

    1. Re:Patents by Anonymous Coward · · Score: 0

      This.
      Uncertainty plus outrageous licensing costs means HEVC is effectively dead.
      You can already see that given how large players are not supporting or are supporting alternate codecs:
          http://aomedia.org/
      Supported by Netflix, Mozilla, Intel, Amazon, Google, MSFT, and Cisco.

    2. Re:Patents by delt0r · · Score: 1

      these codec specs reads like there are 100 engineers each with 2 lawyers and a patent attorney trying to get there little block filter or other obvious optimisation into the spec. Each little "feature" is another patented "invention". It is litterly designed by committee to *be a licensing nightmare*.

      --
      If information wants to be free, why does my internet connection cost so much?
    3. Re:Patents by tlhIngan · · Score: 1

      these codec specs reads like there are 100 engineers each with 2 lawyers and a patent attorney trying to get there little block filter or other obvious optimisation into the spec. Each little "feature" is another patented "invention". It is litterly designed by committee to *be a licensing nightmare*.

      Which is why patent pool organizations were set up. Instead of trying to figure out which of the thousands of people you have to negotiate with, you go to the patent pool, and buy your licenses. No negotiations needed - they're all standard and sold per-unit. Need to make 1000 widgets? Buy 1000 licenses and you're done.

      For the MPEG family of codecs, the pool is the MPEG Licensing Authority, aka MPEG-LA. For h.264, they realized that they needed to get broad adoption which they did by having a ceiling of fees (once you ship N units, each addition unit license costs are "free". Netflix, Apple, Microsoft, Cisco, etc., all do this), free licenses for "free to view" streams (e.g., you can put up ads, and as long as users don't pay directly, it's free). Of course, this lead to wide and broad adoption - since for the most part everyone big paid a lump sum of money for unlimited use (why do you think companies like Microsoft and Apple give away decoders?).

      The big problem is the h.265 patent owners decided they hated this. They saw the widespread adoption of h.264 and that the MPEG-LA wasn't giving them more money. And MPEG-LA was planning on using the same licensing terms to encourage adoption of HEVC (and may help that by increasing fees of h.264 to encourage people to adopt HEVC/h.265).

      Of course, the patent owners disagreed, and formed their own patent pool. Which is fine if everyone did it, but now you have patents only covered by MPEG-LA, and others only covered by the HEVC alliance.

      You can predict the end result - HEVC adoption will probably stall out - there's no reason to use HEVC to justify the extra costs - and yes, that includes sites like YouTube as well which are covered under both the "free" streaming and the license payment cap. (There's no more free streaming of HEVC - while MPEG-LA allows it, the HEVC alliance wants per-stream payment even on free streams).

      So yeah, you'd be hard-pressed to get the same adoption of HEVC as h.264, because the guys involved ignored the fact that the reason h.264 is the dominant codec is because it's easy and cheap to use.

  8. What About Quality? by Assoluto · · Score: 0

    From my personal experience it seems that H.264 Hi10P delivers superior quality compared to H.265. With H.265 the focus appears to have been on reducing bandwidth usage rather than increasing quality, so it's not something that interests me. Besides, H.265 is considerably more processor intensive compared with H.264. I use use the 5 second skip back feature regularly for when I miss a subtitle or if I want to see a bit again. When you skip to a set point in the video the player has to decode everything from the last key frame to that point, which might be up to ten seconds of video. A decent PC can do this near instantaneously with H.264, but with H.265 there is a delay which can feel quite jarring. With disadvantages in terms of quality and performance, I'm not liking H.265. That said, I'm sure video streaming sites will love the bandwidth savings.

  9. Confirms VP9 by Anonymous Coward · · Score: 0

    This test shows what everyone else already knew about (libvpx)VP9. The quality is comparable to HEVC, but the speed is super slow.

    1. Re:Confirms VP9 by Anonymous Coward · · Score: 0

      They used libvpx 1.3.0 in the report. Version 1.4.0 introduced a number of speed improvements for VP9 encoding. You should try it out.

    2. Re:Confirms VP9 by Anonymous Coward · · Score: 0

      yep it goes from super slow to just pathetically slow.

  10. Re:Moscow State? by unixisc · · Score: 1

    There is a Moscow Oblast, which would be an equivalent of state. Although Russia has quite an hierarchy of divisions from federal district to federal subject.

    The other explanation is that 'state' is synonymous w/ government, and that this university is run by the government in Moscow.

  11. Are you kidding me? by Anonymous Coward · · Score: 0

    The technical quality of this report is "unacceptable". How they compare two-pass x264 encoding with single-pass x265. Check the annex with the parameters, even for the unbounded ripping test, one pass is used to x265.

    And they put it at the end so you can realize how much time you wasted on this report when you reach the end....

    1. Re:Are you kidding me? by Anonymous Coward · · Score: 0

      No kidding. Encoding times and resource usage varies wildly just by changing parameters on the same encoder. Any comparison between encoders will be highly subjective due to the observer's skewed options.

    2. Re:Are you kidding me? by Anonymous Coward · · Score: 0

      To quote the test objectives, emphasis added:

      The main goal of this report is the presentation of a comparative evaluation of the quality of new HEVC codecs and codecs of other standards using objective measures of assessment. The comparison was done using settings provided by the developers of each codec. Nevertheless, we required all presets to satisfy minimum speed requirement on the particular use case. The main task of the comparison is to analyze different encoders for the task of transcoding video—e.g., compressing video for personal use.

      So I'd say those were the best possible settings for x265 in the context of the comparison.

  12. how doest it compare to daala ? by Anonymous Coward · · Score: 0

    how doest it compare to daala ?

    1. Re:how doest it compare to daala ? by Anonymous Coward · · Score: 1

      Are We Compressed Yet uses some objective metrics to compare the current Daala development to x264, x265, VP9, and Thor. Although bear in mind that objective metrics aren't perfect and don't always tell the whole story. See pages 27 and 28 of these NetVC presentation slides. To me the Daala encoded image looks better and captures more detail, but it scores worse on the objective metric.

  13. Re:Moscow State? by Anonymous Coward · · Score: 0

    since when does an American only league constitute a world series!

  14. Re:Moscow State? by bloodhawk · · Score: 1

    So much ignorance in your post. An oblast is roughly the equivalent of a state. There is a Moscow Oblast. probably should try using google before making yourself look like an idiot.

  15. Can you give build 1018 a try? by Anonymous Coward · · Score: 0

    ... with build #1016 (which includes FFMPEG 2.8.1) I was getting smooth playback and averaging around 60% CPU on both cores ...

    I am interested to know how the thing fairs with the newest version, build # 1018

    Can you give it a try and then share with us your experience?

    Thanks !

  16. This Was Very Helpful by Anonymous Coward · · Score: 0

    This was very helpful. Thank you!

    Based on your numbers, I expect that if I add a GTX960 I will get a 3X increase in my encoding rate for 2160p HEVC with -vcodec
    nvenc_h265.

    I would have liked a Quicksync solutoin better, for my i7-4790K. But, I'll still be happy with the GTX960 option.

    Thanks again.

    1. Re: This Was Very Helpful by Anonymous Coward · · Score: 0

      Hi all, we develop encoders based on NVENC and FFmpeg and we can go double 4K@60fps easily with GTX Titan X on HEVC live stream. GTX 980 can make a single 4K@60fpa HEVC and 960 can make 4K@30fps HEVC. We have all cards and did the tests. Piko.tv / ali@kizil.com