MPEG-4 AVC/H.264 decoder comparison

[BetaBoy]

yeah... its in the bug tracker... thx

[siddharthagandhi]

Could someone answer my question about what benefits I would recieve from getting a specialized decoder on my high-end system?

[Caroliano]

A faster decoder use less CPU cycles. Then, if you want to run another process that needs a lot of CPU, you will have more cycles left to it, then it can be faster/better. Like if you want to encode while viewing a 1080p movie.

You will have no quality gain in the decoding, as all AVC decoders that are spec compilant produce the same quality. Of course, you can use the spare cycles to run Lanczos Resize instead of Bilinear, and other things like that, for improve quality in playback.

[siddharthagandhi]

Then that basically means I don't need it. I'm not just going to give out money unless I don't need to. I don't encode a lot. When encoding AVC at a bitrate of 16 mbps in Nero Recode I can watch the WMV HD 1080p trailers with no lag. So I doubt that an AVC file would overload my system.

[CruNcher]

And offcourse you will save energy on laptops and mobile devices for example longer battery life :)
and then Energy saving means less Temperature so your CPU will also last longer ;)

[siddharthagandhi]

I've got a good fan so I don't have to worry about that.

[IgorC]

Average CPU speed isn't that fast. And only with new CoreAVC decoder thousands of people were enable to play their H.264 DVD/HD videos.
Before even 3.2 ghz CPU wasn't enable to decode in real time HD 1080p. Today it's possible with CoreAVC.
And not everybody has 3.2 CPU.

Thousands people said their "Thank you" to CoreAVC devs.

[shon3i]

I have new Cyberlink H264 decoder from PowerDVD 7 but seems to be very slow decoder.

[Shinjite]

Yup, so far CoreAVC is still the fastest video decoder out there for h264

[Nick [D]vB]

If only it had MBAFF support.... :D

[BetaBoy]

Not sure if I stated our schedule... but unlike what I had stated earlier we are adding HE/LC/SBR support to CoreAAC v.2.0 now rather then later. This has effected our schedule somewhat but we are still on track to add MBAFF to CoreAVC Professional and Enterprise Editions soon after CorePlayer Mobile 1.0 and BetaPlayer GPL are released next month.

[Nick [D]vB]

Hello again,

Sorry to keep on about this but I’m going to give it one last shot. I appreciate you have set an internal development schedule according to your available resources and perceived demands of the marketplace but I just won’t to make sure you are fully aware of the current situation in the UK as I feel there is a potentially huge and, as of yet, untapped market for you here.

I’m sure you are aware the BBC have just started an HD trail across all platforms, it will show a wide range of programming but without question the biggest attraction will be the football World Cup which starts in just under a month. There has been a lot of media coverage of the trail and specifically the fact the major providers have massively underestimated demand for equipment that can receive it. Watch the short national News segments here for the background:

http://news.bbc.co.uk/1/hi/programmes/breakfast/4752017.stm

Basically the only cable provider offering the service, called TeleWest, is currently only available to a small number of homes in the UK. The other major player is $ky satellite, it has enough kit but is having trouble training enough installers in time, and now can not install any systems until after the World Cup has finished. :mad:

If I’m boarding you just bear with me a minute this next bit is crucially important. The BBC had also started a DVB-T trail which would let a large part of London receive the same HD programs through a normal TV aerial. BUT the BBC has chosen to run this as a closed trail [due to bandwidth limitations and other “political issues” with the regulator OfCom] and so has only supplied 500 set-top decoders! The service is FTA and unencrypted but there are no other commercially available AVC DVB-T decoders that can be used to watch the service. :confused:

Because of this a PC based solution seems to be the ONLY option for many people wanting to watch the World Cup in HD this summer, and there are plenty of them! In fact it would be hard to over-estimate the demand here at the moment, I know hundreds of people currently looking for a real-time MBAFF decoding solution in time for the World Cup. According to tests I’ve done with the Elecard and Cyberlink decoders they are just not able to provide acceptable performance even on ultra high-end dual core A64FX based system with ATI X1900’s [using GPU acceleration]. I don’t know about NVIDIA’s acceleration performance but most report it is not working with most products in the latest drivers.

I am sure if you are able to provide a solution in time for the Worlds Cup [even a fuctioning beta release] you would have a massive number of customers, but the timing is critical, with your current development schedule MBAFF support would not be available until well after the World Cup has stated, I don’t want to jinks the team but should England be knocked out of the tournament in the early stages a large part of this potential market would just evaporate over–night. Let’s just hope they are still hanging in there by the time you’ve got something ready! Well I’ve said my piece. I am not trying to tell you how to run your business; I just wanted to make sure you were fully aware of this opportunity. Keep up the good work guys – just go a bit faster… :D

Thanks,
Nick

[f@chance]

I add my penny/nickels worth of opinion. What makes it even worst is, that the Germans started broadcasting in HD since November and the Worldcup is being filmend and aired in Germany in HD. Now if you know even a bit about Brits and Germans than you know that there must not be a HD gap between them.

There is enough disposible cash in the UK that in the event of a semi final replay of the 1990 game between Germany and England, that they will go through any lenth to watch that in HD format, regardless of the cost. Businesses are heavily advertising for new Worldcup gear and the buzz is contagious.

In the last worldcup during the England, Argentina game business came to a grinding halt and traffic congested roads looked deserted.

Just my bit as a German, reared in the US and having lived for the last 10 years in London. For the UK there is no other important event then the Worldcup, esecially since it has been 40 years since they won it and every 4 years the press reports, this time we have a chance to win it.

[Doom9]

kick BBC where it hurts and demand a progressive signal. Interlacing is beyond outdated technology and there should be laws against producing such crap.
And we can consider ourselves lucky that they'll broadcast the whole thing in widescreen this time.. it has only been like half a decade since widescreen TVs started outselling 4:3 models.

[Sulik]

I disagree.
At the same bitrate, interlacing will produce better quality than progressive (assuming proper deinterlacing). It inherently has full vertical resolution when there is no motion, and half-resolution but double the frame rate with fast motion.

I do agree that things would be much simpler if everything was progressive, though.

[SeeMoreDigital]

kick BBC where it hurts and demand a progressive signal. Interlacing is beyond outdated technology and there should be laws against producing such crap.
And we can consider ourselves lucky that they'll broadcast the whole thing in widescreen this time.. it has only been like half a decade since widescreen TVs started outselling 4:3 models.At the present time, there's only a handful of TV channels being broadcast in HD via Sky-HD. All of which are encrypted. All of which appear to be favouring 1080i50 transmission protocol - So no 720p50 yet :scared:

If the BBC ever get around to broadcasting an full-time HDTV service via DVB-S2 (as apposed to their useless DVB-T trial service in London) they'll still be the only UK TV provider broadcasting in the clear.

Sadly, their AVC samples are just that, offering no clear indication as to whether 720p or 1080i will be used in the long term :(

[Wilbert]

And we can consider ourselves lucky that they'll broadcast the whole thing in widescreen this time.. it has only been like half a decade since widescreen TVs started outselling 4:3 models.
Over here they are morons. On one hand they broadcast it in HD, but on the other hand the analog broadcasting is not even in widescreen. The official reason is that people with normal TVs don't like the horizontal black bars nor a distorted image.

It's funny though. 9 out of every 10 people I know with a widescreen TV, prefer a distorted image over vertical black bars when watching fullscreen broadcasts :)

[Nick [D]vB]

Now if you know even a bit about Brits and Germans than you know that there must not be a HD gap between them.
LOL, leave me out it, I'm from Holland so me and Willbert are netural! :D

(as apposed to their useless DVB-T trial service in London)
Not that useless if you don't live in a cable area and can't put a dish up. I know lots of people in the area looking for a way to get the DVB-T trial. Remember Channel 4 / Flim4 and Channel 5 will be adding streams to the HD trial soon, they will be FTA on DVB-T but I'm pretty sure they'll have to be NDS encrytped on satellite?



BTW, have you noticed they are using 1440 X 1088 resolution? What's that all about? Any ideas why they're not using standard 1920 X 1088 [especially @20Mbps] ???

It's funny though. 9 out of every 10 people I know with a widescreen TV, prefer a distorted image over vertical black bars when watching fullscreen broadcasts

I'm not sure what that says about the Dutch, maybe if you have paid for the whole screen you should use it all! :D

[f@chance]

Just to add some additional information. Today was the first day of Sky HD installation and some users were pleasantly surprised others had a rude awakening.

http://news.bbc.co.uk/2/hi/entertainment/5003728.stm

Sky has delayed the installation of high-definition TV (HDTV) in 17,000 homes because it's supplier has failed to deliver enough set-top boxes.

Anyone requesting installation on Monday, launch day, will be told that they must wait until early August, meaning they will miss the World Cup - which begins on 9 June - in HD.

What can I say there are a host of Londoners that have their eyes peeled on CoreAVC with MBAFF. For some it will be let the games begin for others it will be lets head to the pub.

[dimzon]

Just suggestion
Why CoreAVC doesn't included in test? Maybe it's time to update
I strongly prefer to vie results graph first (before preparation/explanation)

[Nil Einne]

Am I right that CoreAVC is faster then with Nvidia's PureVideo or any other codecs which support Nvidia's hardware acceleration (if there are any other codecs)? I've seem some reports but they weren't particularly clear. And this is also true for ATI's AVIVO?

If so, it'll be interesting to see how CoreAVC Pro performs when it has support for GPU acceleration.

[Ice =A=]

@Nil Einne:
Have a look here (http://www.behardware.com/news/8117/coreavc-stronger-than-avivo-purevideo.html)! (This has allready been posted somewhere in this forum by the way.)
Short summary: CoreAVC is very fast, fastest software decoder by far, however at the moment (!) there is one combination which is a little faster, nameley PowerDVD with an nVidia card.

[Nil Einne]

Actually, although I didn't search, I did come across this independently although I just skimmed through it. It was why I thought CoreAVC was better then the rest even with hardware acceleration. However even the article itself admits that it isn't that good a test. Specificially "To finish, we need to emphasize that it is a just a preview based on trailers". Also no clear mention of the methodology for measuing CPU utilisation and for that matter whether they repeated their results. I admit, I'm always very skeptical when I see tests that don't (to me) explain how whatever they were measuing was measured and whether they attempts to account for variation due to other factors.

It would be interesting to see a mode detailed test akin to what was done by bond. Also, it would be good to see a test with any released & easily available versions of codecs, even if these are designed for Nvidia or ATI.

But actually now that I look more carefully, I realised in fact you're right. With the Cyberlink codec and 7900GTX (and therefore, probably the 6600GT which I have since I believe the difference between the H.264 acceleration part is minimal, or am I wrong?) is better then the CoreAVC, although this is not with CoreAVC 1.1 of course. Cyberlink's acceleration engine is clearly done much better then the Nero one and of course CoreAVC beats them hands down in pure software. So I guess I reach the same conclusion. It'll be interesting to see what they can do with GPU acceleration and for that matter what the others can do to catch up. Especially the open source guys...

[foxyshadis]

The AVC decoding speed strongly depends on the overall speed of the card as well, both cpu and memory. (6800 is broken, but 6600 is not.) 7900GTX gives a large boost in decoding speeds, or rather, how high a resolution and how much extra you can make it do. So someone else's benchmarks aren't of much use to you unless you have the same card and processor.

Of course, benchmarks are always welcome if you can come up with a good one.

[Revgen]

@Nil Einne:
Have a look here (http://www.behardware.com/news/8117/coreavc-stronger-than-avivo-purevideo.html)! (This has allready been posted somewhere in this forum by the way.)
Short summary: CoreAVC is very fast, fastest software decoder by far, however at the moment (!) there is one combination which is a little faster, nameley PowerDVD with an nVidia card.

CoreAVC will have GPU acceleration in the future and I'm pretty sure it'll beat whatever Cyberlink has.

[Ice =A=]

Very little doubt about that! :)

@foxyshadis:
I'm also very sceptical about someone elses benchmarks (in Germany we have a saying: Only believe in a statistics if you have faked it yourself...), but it seems that's all we've got at the moment.

[CruNcher]

cheating would bring them nothing Ice =A= everyone can check those results and if they were faked, we would allready knew it (and the whole internet) :)

[Nil Einne]

It's not so much cheating, but sloppy benchmarking. For example, so often benchmarkers don't bother to run any repetitions and only test one or two things. Then they say, we have proven XYZ and I say bullshit you haven't shown shit.

From my own benchmarking I've seen sometimes results can be fairly variable. Whenever I'm actually comparing something I always take averages and do standard deviation. It's stupid to claim there is a difference if the difference is well within normal variation.

It's also a bit of a fallacy to assume that cheating or sloppy benchmarking will always be detected. So far, I haven't seen any other public benchmarks comparing with and without GPU. Of course, I'm not saying others haven't run private benchmarks which concured with the published results so they didn't say anything however we have to reason to assume people have. (Looking at science should give you a clue. The standards here are a lot higher then with benchmarking and benchmarkers but people still get away with cheating and sloppy work. Of course, it is usually a lot hard to repeat a scientific experiment of some kind then a benchmark and there is possibly a smaller base doing it but it does show the fallacy of assuming something will be detected. Eternal vigilance and caution is always in order especially when only one site has published a result)

I'm aware Core are developing hardware acceleration support and I'm sure it'll beat the hell out of what others have used however this still doesn't answer the question of whether GPU accelerated Cyberlink or whatever is faster then non-GPU accelerated CoreAVC. Obviously it'll depend on what your precise specifications are and what file your using and indeed that was one of the points I was making.

But with quality benchmarking which IMHO should include multiple variables (different GPUs, different CPUs etc) you can get a very good idea even without benchmarking yourself. BTW, there are actually 3 issues here. One is which one uses less CPU. Another one is which one is faster (this is not the same thing as less CPU when the GPU is involved). Finally power usage is also another issue which many will be interested in. Generally speaking the first one matters most. If your decoding something for realtime playback, provided your decoder can keep up you usually don't care how fast your decoder can decode. Lower CPU usage however might matter if you have background tasks that will either use or might use your CPU (e.g. on a PVR). GPU usage is generally not so important although with ATI's stream for folding@home I guess some people will care. However CPU usage is very tricky to properly benchmark (whereas decoding time is generally a lot easier).

Power usage is the dark horse here. Some people will care a lot, some not at all and some in between. However power usage is one thing that has so many variables that IMHO it is very difficult to reliably benchmark unless there is a large difference.

N.B. I'm not intending to offend any benchmarkers. I'm just airing my POV. Also, I primarily rant about benchmarks published on websites and I'm mainly talking about comparing CPUs, GPUs, fans and all that kind of stuff rather then comparing codecs although the same principles apply.

P.S. To be honest I don't really care that much either way. It was just something that I was wondering at one stage so decided to check it out. My PVR has a Radeon 9000 so no GPU acceleration and my 6600GT is on my A64 2.4ghz which is fast enough for my needs that it doesn't really matter. I don't use HD and in fact I've only ever looked at 1 or 2 h.264 videos (with VLC...). In fact I'm more interested whether the open source guys can respond to this challenge then anything else but thats a different issue.

[Hans Ohlo]

one thought i have at the moment: shouldn't every h264 decoder output a bitperfect identical picture as requested by the spec? it seems not so, because every decoder i tested shows other characteristics: cyberlink is a bit unsharp, coreavc has big problems decoding 1080paff streams (restless picture, broken horizontal lines, color level bug), mainconcept seems ok but seems unsharp against ffdshow...

how comes all this? and what decoder is nearest to the spec?

[KoD]

Well, you have the h264 reference decoder (JM (http://iphome.hhi.de/suehring/tml/)). Compare the frames with what that one outputs. It's an app that you have to compile yourself and it's obviously not meant for real-time playback.

[BetaBoy]

coreavc has big problems decoding 1080paff streams (restless picture, broken horizontal lines, color level bug)The next release fixes all of the known issue such as you describe. As far as being compliant... outside of the JM reference I would say CoreAVC and the Ateme decoders are the most precise... but take that with a grain of salt.

[ViRGE]

So I've been reading through this thread and it seems rather dated, so I hope someone doesn't mind my noobish questions.

I have a Radeon X1900XTX and I'm looking to find a nice H.264 decoder to use with it that supports GPU-accelerated decoding. I've noticed that there's a specific combination that seems to be recommended for GeForce users, but I haven't seen such a thing for Radeon users. Is there any 1 setup that's better than all others? I'm not aware of any codecs besides those from Cyberlink and Intervideo supporting GPU-acceleration at this time(am I missing anything?), so this seems to be a one-or-the-other thing.

Also, how well do the above support non-commercial content? I'm asking mainly so I can play stuff encoded with x264, which seems to be the popular choice for non-commercial stuff. I suppose in the same vein, do both work as DirectShow codecs so that I can use them with Media Player Classic?

[Atak_Snajpera]

I would be very happy if libavcodec started to use multi-cores to decode h264. Even my overclocked C2D E6300@2,8ghz is too slow to decode 1080p@25fps (15mbit/s). Friend of mine has C2D@3,6ghz so we will check if is fast enough... :)

[Theliel]

umm you are sure?? with latest ffdshow I can play h264 1080p perfectly with my old P4 3.4 HT, and the CPU load are about 70% max 40% min.

[Atak_Snajpera]

Your cpu maybe old but still has higher clock (21% faster). libavcodec is optimized (I assume) for vector units like SSE and SSE2. SSE unit is able to add 4 single precision floating-point numbers (32bit) using only one clock (SSE2 - 2 double precision ). This simple example shows that if you have higher clock you can have higher MFLOP rating. BTW check my screenshot

http://img359.imageshack.us/img359/7876/xmen3ji6.th.jpg (http://img359.imageshack.us/my.php?image=xmen3ji6.jpg)

[foxyshadis]

Bitrate is more important than resolution. 1080p @ 1.5Mbps is vastly much easier to decode than 1080p @ 15Mbps. Try a little of the latter and see if your P4 chokes.

btw, mflop doesn't matter for video, integer performance is what constrains it (when memory speed etc is ignored). Even if it did, it still depends on what you're talking about; a single Conroe core has a lower mflop rating for x87, but higher for floating-point SSE2, which executes twice the speed per clock of P4.

[KornX]

has sb tested nvidia purevideo HD?

KornX

[CruNcher]

intensively on the Geforce 7 yes for Mpeg-2, H.264 and not so intense for VC-1 Playback, also in terms of Visual Quality of the Motion (Pixel) Adaptive Deinterlacer by Nvidia (not every Driver release gives good performance here) but the Architecture of the new Forceware 100 series driver is not the same anymore then the old 95/96/97 releases internaly it has changed for the Geforce 8 you can use it for 3D on the Geforce 7 but Video (PureVideo SD/HD) quality is not as good as with the Geforce 7 Driver (at least under XP maybe that changes in July when Purevideo is gonna work for XP on the Geforce 8 then) for Geforce 7 based cards (Motion Adaptive Deinterlacing doesn't work anymore with the Forceware 100 series Driver on Geforce 7 cards currently under XP not tested under Vista, so beaware).

[SoreGums]

So does the CoreAVC codec work under Vista?

I've paid for it but since upgrading to Vista I haven't been able to install it due to it being activated... They don't reply to emails either which sucks since you have to pay real money for this thing...

I've installed the Vista codec (v444) pack and ffdhow is installed and taking care of x264 but the playback is not very smooth at all.
Turned PostProcessing off and that helped but its still not smooth when lots is happening in the frame.

Should I just buy PowerDVD?

[BetaBoy]

Visit http://support.corecodec.com and submit a support ticket and the staff will reset your account to allow the upgrade.

[SoreGums]

Visit http://support.corecodec.com and submit a support ticket and the staff will reset your account to allow the upgrade.
I've tried doing that - but it just says invalid login/password . . . . - So I went back to the email and never got a response.
Same username as here

[cyberloner]

does anyone here can really use hardware to decode with dxva to play video such as (h264 / vc1/ qt mov) or bluray / hddvd
powerdvd ultra or what ever codec maybe?
i try all failure.. even with latest powerdvd ultra patch....

[BetaBoy]

I've tried doing that - but it just says invalid login/password . . . . - So I went back to the email and never got a response.
Same username as here

Your support account is diff then the registered account, so you need to register to submit a ticket. We are migrating to OpenID later this year for all our sites/services. Till then its a seperate process for each service (purchase, support, etc.).

[land]

Hi, I think it will so interesting a new test with the new versions of the h264 decoders.
One similar to the firt post in this thread.
The decoders could be Ateme, livacodec, elecard, moonlight, mainconcept, VSS H264 and maybe quicktime.
I would like to know the sopported options of each one and its speed.
I don't know how to do the test, that's why I write this post.
Is someone interested in do it?

[Dark Shikari]

Hi, I think it will so interesting a new test with the new versions of the h264 decoders.
One similar to the firt post in this thread.
The decoders could be Ateme, livacodec, elecard, moonlight, mainconcept, VSS H264 and maybe quicktime.
I would like to know the sopported options of each one and its speed.
I don't know how to do the test, that's why I write this post.
Is someone interested in do it?
Use TimeCodec. And make sure you get the latest FFDShow-Tryouts--I hear someone got some multithreaded decoding support for H.264 in it.

[Sagekilla]

Use TimeCodec. And make sure you get the latest FFDShow-Tryouts--I hear someone got some multithreaded decoding support for H.264 in it.

IIRC, Aku said it worked on a per-slice basis and not per-frame, and since x264 doesn't encode videos on a per slice basis it receives no benefits. It was something like that, I'm a bit fuzzy on details but Aku basically said ffdshow won't offer multithreaded acceleration for the newer x264 videos.

[Dark Shikari]

IIRC, Aku said it worked on a per-slice basis and not per-frame, and since x264 doesn't encode videos on a per slice basis it receives no benefits. It was something like that, I'm a bit fuzzy on details but Aku basically said ffdshow won't offer multithreaded acceleration for the newer x264 videos.I also heard (IIRC) there was some work after the slice-based parallelism that added some non-slice-based support--its all experimental of course.

[Manao]

Its all experimental of course.It was a patch allowing the threading of the bitstream decoding in parallel of frame reconstruction & deblocking. It was working only for non-mbaff videos, and I don't think the patch can be cleanly applied now that PAff has been implemented.

But there seems to be some work on frame level parallelism.

[COOL-GOOD]

:goodpost:Ateme MPEG-4 AVC H.264 Encoder PC Suite

i believed that a program "Ateme MPEG-4 AVC H.264 Encoder PC Suite" is good nothing like for Video Converter after the experiment of many of other programs...

> MPEG-4 AVC / H.264 Baseline, Main, High Profile
> From QCIF to Full HD 1080i
> Advanced audio codecs
> Up to 7 audio tracks encoding
> DVB subtitles and DVB Teletext
> Professional pre-processing
> Batch processing and job manager
> Both SD and HD full support in the same product
> Reduce bitrate up to 50% compared to MPEG-2 and preserve the same high quality
> Optimized architecture for multi-processor platforms to achieve real-time encoding
> Easy automation to run 24/7 processes
> Fully interoperable with major Set Top Boxes and VOD platforms
--------------------------
VIDEO ENCODERS

> H.264 BP and MP
> MPEG-4 SP
> JPEG/MJPEG

VIDEO DECODERS

> H.264 BP & MP
> MPEG-4 SP & ASP
> Xvid & similar codecs
> MPEG-2 MP@ML
> WMV/ VC1
> JPEG/MJPEG
> Enhanced JPEG for Consumer Electronics aaplications
AUDIO COMPRESSION

AUDIO ENCODERS

> MP3
> AAC
> MPEG-1 Layer 2
> ADPMC

AUDIO DECODERS

> AAC LC & HE
> AC3
> MP3
> WMA
> ADPMC
PRE-/POST-PROCESSING

> Resizer
> De-interlacer
> Dithering
> RGB/YUV conversion
TOOLS

> Streaming protocols
> File reader/writer
> A/V Sync
For more informations:
http://www.ateme.com/TEC_codecs.php

@edit by admin: Pirated tools have no place on this board.

[Sagekilla]

I also heard (IIRC) there was some work after the slice-based parallelism that added some non-slice-based support--its all experimental of course.

Shame.. I'd (And I imagine quite a lot of people) love to see a multithreaded decoder that could take advantage of my dual core processor. Until then, CoreAVC is doing a magnificent job at decoding even 1080p video.