Once again we've come to the last page of a codec comparison and are asking the dreaded question "which is best?". Actually, if you're a member of my forum you know better than to ask that questions but I'll try to answer it anyway. But be aware that you are not only entitled to see things differently, you might actually see things differently and can in all honesty disagree with my findings (with the proper respect though and some reason.. it's hard not to notice the detail difference between RV10 and the NeroDigital AVC codecs but you might value this differently). After all, you don't have my eyes and I don't have yours.

Let's start with 3ivX. The codec has once again been improved. Its rate control is accurate, it now offers an almost complete MPEG-4 ASP featureset, and it is quality wise up to the level with DivX, which seems to have been the goal.

I was once again not impressed with the progress DivX has made. It's nice to have additional features available, but quality wise I just don't see one year of development. Plus, if you spend as much time using settings that are called insane by the codec, I'd expect to see something. Though, this time there were no problems with Futurama, something which was a problem in the last test.

HDX4 left a mixed impression. There were no apparent issues in Futurama, and up until the night scene, I kinda liked it in SPR as well. Perhaps the effect of the grain reconstruction mechanism is a matter of taste and perhaps it'll become less disturbing as development progresses. But, for an upstart codec, it definitely rated much better in the first attempt than many other codec.

Ateme's NeroDigital AVC codecs really impressed me. The Main Profile encoder is already available to every Recode user, and thus a very good AVC implementation is already publicly available. Considering that AVC was only ratified in spring 2003 (and only became an ISO standard in December 2003), and that 1.5 years later we already have a fast encoder that delivers good quality, that's quite a step ahead from how long it took MPEG-2 and MPEG-4 to make that step. The High Profile codec left a very slightly better impression, but I cannot shake the feeling that I just expected more. But, not all of the features of the High Profile have been integrated yet, and considering that the profile has just been ratified in November 04, I guess I cannot expect too much right now (though still, it's the best codec I've tested).

RV10, where should I begin? I know some people really like it, and it leaves a visually pleasing impression, but when you compare the results to the source, it leaves to be desired. And in my opinion, the progress since the last test has been less than stellar.

VP6 has not been under heavy development for months, but since the last comparison it has still been visible improved. It managed to beat almost all MPEG-4 ASP contenders, which is quite an achievement for a small company like On2. Now if that darned rate control would not give me oversized files and if the codec was faster.. Also, the considerable oversize obviously gave VP6 a bit more bitrate to work with, so I'd feel much more comfortable if the size had been on target in the two full lenght movies and if the results were still the same.

Videosoft's AVC encoder exhibited perfectly fixable problems that many codecs have head in their early age. But I think the codec has some definite potential that can eventually be exploited.

WMV9: With WMV9 being standardized as VC1, we'll probably not see more features in it, but having a standardized bitstream and featureset doesn't mean there cannot be any progress. It's just that I have not really seen it since the beta. But, detail wise some things have changed and I recall rating WMV9 in the "less details" area, which it no longer is (if you look beyond the problems that WMV9 had on my machine). But, if I were a broadcaster or potential user of HD DVD/ Blu-Ray, seeing the result of WMV9 and ateme's AVC codecs, I'd definitely go for the AVC standard.

Finally, XviD, one year after taking the crown, had to give it back. It would've won again, if it were not for ateme's AVC codecs. So, if you make DVD backups now that need to work on a standalone or slower machine, XviD is still a very good option, but I guess we'll see AVC capable decoder chips in 2005.

Standalone playback

Currently, DivX and XviD are your best solutions if you aim to make your backups work on an MPEG-4 enabled DVD player. 3ivX and HDX4 should pose no problems but it is to be assumed that most players will not recognize the FourCC code used, so keep in mind to change it. Some players only handle DivX, but can be made to support XviD using the same trick (though such players are rare, many player makers specifically advertise XviD playback as a feature - which shows that an open source project, with no advertising budget whatsoever, can clearly make a significant impact in today's world).

WMV9 hardware playback is also taking off. There's currently one chipset maker that already has a working offering for WMV9 playback, and there will be more in the future. Personally I'm still a bit skeptical if WMV9 is really going to be used (I guess it depends on how fast professional hardware AVC encoding devices are out for large scale broadcast and disc production becomes available), but I suppose we'll see more and more WMV9 enabled hardware. But will it become as popular as MPEG-4 ASP hardware? I think it stands and falls with WMV9s success or failure in the high definition DVD area.

When VP6 is concerned, it should be an option in China's EVD format, but that has so far failed to make a real impact, and it is doubtful that the format will make it outside of China. And even in China, there's already several new formats.

There are no available nor planned players for RealVideo.

We'll see the first AVC enabled DVD players in 2005 and I expect a somewhat faster takeoff than we've seen with ASP players. But for the future, I think AVC will definitely be a valid option.

Best quality per fps

I decided to introduce this part when impatiently sitting in front of my two encoding machines (2.8 and 3.5 GHZ CPUs respectively) and wishing my encoding sessions would finally be done.

Looking at the encoding speed table, this was an easy pick: XviD clearly delivers the best quality per FPS and shows that high speed is not detrimental to quality at all. Also, ateme's Main Profile encoder delivered a good 31.40 fps, which is very respectable for an AVC codec, and thus it earned the 2nd place in this category.

Besides the codecs

Today, some codecs no longer come as a simple VFW encoder and DirectShow decoder solution:

3ivX has an entire filter suite. Besides the MPEG-4 video encoder there is an AAC Low complexity audio encoding filter, an MP4 file muxer (so that you can turn AAC and MPEG-4 video into a specs compliant MP4 file), and playback filters to handle MP4, multichannel AAC audio (even High Efficiency AAC) and generic MPEG-4 video.

DivX ships with a media player and an application to tweak the bitrate distribution of the video codec, called EKG. DivXNetworks also offers a one click DVD encoding tool called Dr. DivX.

HDX4 is not available yet, but will also show up as part of a tools suite. Its decoding filters can also handle NeroDigital specific features like chapters.

NeroDigital is not available as regular VFW codec, nor as the commandline utility I've used for these tests. Rather, it comes integrated into Nero's Recode one click DVD backup suite, which not only features NeroDigital output but can also handle DVD-R output. Recode's NeroDigital output comes as an MP4 file with AAC (High Efficiency if you like, both 2ch and 5.1ch is possible), subtitles and chapters (though the latter two require that you use Nero's Showtime player, you won't get those features in any other player). What's still lacking is a suitable MP4 editing tool though (it should eventually be possible to edit MP4 files in NeroVision Express, but I guess I'm just waiting for an MP4 capable VirtualDub (and considering where we're moving with AVC encoders it should be a DirectShow based tool).

RealNetworks not only offers their own video codec, but also has both 2ch and 5.1ch audio codecs, as well as an AAC audio encoder. The RealMedia container also supports streaming, subtitles and chapters. Editing is problematic though as RV10 editing isn't a feature of mainstream video editing applications. It is also only possible to use RV10 outside of the RM container when using Matroska.

WMV9 is also part of the WindowsMedia tools suite. There is WindowsMedia Audio, an audio codec that can handle up to 7.1 channels, and Microsoft's WMV container, which can also contain chapters if I'm not mistaken. Editability of the WMV container is better than with certain other containers, but still cannot keep up with AVI.

At this point, all participants support AR signaling in encoder and decoder regardless of container, so it is possible to encode a movie anamorphically (as it's done with 16:9 movies on DVD).

Future outlook

3ivX will continue their ASP line with the upcoming release 5.0 and even beyond that. At the same time, they're also working on an AVC encoder, which hopefully will be ready for the next comparison.

DivX Fusion had a version number of 5.9, so I think once the beta phase ends, this will be available as DivX6. Beyond that, I have no information on future plans for DivX.

HDX4 is yet to be released, but its maker Jomigo is already looking into AVC and we can expect to see an AVC solution from them as well.

The NeroDigital Main Profile encoder I tested will be available as part of the next Recode release. ateme will continue to improve both Main Profile encoder as well as their ASP encoder. And obviously, work on the High Profile encoder has just begun. In the future, we'll see more features used than just the 8x8 transform already present in the encoder's current state. In the future we'll definitely see custom scale matrices and of course the film grain feature. We have already seen some NeroDigital ASP capable players, and Nero is actively pushing those (but things apparently took time as this program was announced a year ago and there are very few players that support NeroDigital at this point). One reason for this might also be that Nero keeps the specs of their certification under NDA. How is customer suppose to know what he/she is buying if you don't know what a certification involves? DivXNetworks does this more intelligently and reveals the details of their certification to anybody (although I'm still waiting for a full ASP profile ;) An NeroDigital AVC certification is also forthcoming.

On2 will release VP7, probably some time in 2005. Considering the results of VP6, we can probably expect good results from VP7 (outrageous claims by their marketing department notwithstanding ;)

Considering that WMV9 is currently undergoing standardization, and is part of the HD DVD and Blu-ray specs, we can most likely see most of the action in the hardware department. Considering the improvements since WMV9's first release, I'd expect a slow progress.

XviD 1.1 is bound to be released soon, but that's not the end of it. We'll see more improvements coming in 2005, and the developers are also looking closely into grain preservation mechanisms.

I have no information on the future of Real's codec and the VSS one.


In general, I think we'll see a lot of activity in the AVC department in 2005. We'll see hardware players and codecs by more and more companies. Considering it took MPEG-2 9 years since its ratification to really get to the level we are today (it was ratified in 92 and when it comes to progressive content I think we've reached a peak in 2001 with CCE 2.5 and the Robshot method), and that MPEG-4 ASP codecs still make progress (MPEG-4 was ratified in 1997), I'd expect that AVC still has lots of places to go. I also expect development to be quicker than MPEG-2 and MPEG-4, particularly because both broadcast industry and Hollywood is looking into this format for more efficient storage of TV programming and movies.

Now if you have any questions, please have a look at the FAQ.

This document was last updated on December 28, 2004