v5.11 speed

[Doom9]

I know these discussions are almost always pointless but in case of mine you get all the details so that it is comparable: The scenario: Codec comparison, Matrix using the same setup as last time (http://www.doom9.org/codecs-103-1.htm).
Settings: no profile, B-frames, slow mode and slow psy settings.
I've always had DivX5 for a fast codec but the numbers I'm getting now are extremely low so I'm wondering if I'm not the only one getting low values. As you can see following the link above, I used to get 40 fps, now I sometimes don't even have to use two digits for the speed anymore. Note that this is on the same hardware as in the last comparison as well.

[temporance]

Hi Doom9,

Yes, 5.1.1 is slow in slow[est] mode. IMHO(as with xvid's slowest modes) these modes really are for people with a fast PC left to encode overnight. The amount of extra time invested probably only yields a very small perceptual improvement unless you're really pushing bitrate down.

One thing that might be interesting is to plot a scatter graph:

Overall PSNR (dB)
^
|
|
|
|
+-------------> log10(total encoding time)

This will show the diminishing returns of increased encoding times (and be a useful way of placing fast codecs like Nero Digital).

Btw, (psy fast is much much faster that psy slow, but the characteristics of the artifacts are different). If you want to compare PSNR, don't use either PSY mode as (same as other codecs' PVE) these features will, by design, worsen PSNR figures.

[Soulhunter]

XviD (http://forum.doom9.org/showthread.php?s=&postid=415627#post415627) vs DivX (http://forum.doom9.org/showthread.php?s=&postid=412191#post412191)

PSNR wise XviD got better result at less than 1/3 the time than DivX !!!

Bye

[SeeMoreDigital]

Hi Doom9,

When you did your last DivX tests you used DivX5.0.5 at 'slowest' mode. When using DivX5.1.1 this equates to the 'standard' mode!

That said, the tests I have done using DivX5.1.1, show that as soon as you put a tick in that BiDi box, the encode speed drops like a rock. Something that was not so apparent with DivX5.0.5.

Cheers

[Nazgul]

Originally posted by Doom9
Settings: no profile, B-frames, slow mode and slow psy settings.
I've always had DivX5 for a fast codec but the numbers I'm getting now are extremely low so I'm wondering if I'm not the only one getting low values. As you can see following the link above, I used to get 40 fps, now I sometimes don't even have to use two digits for the speed anymore. Note that this is on the same hardware as in the last comparison as well.

If you want to compare to Divx5.05, use Standard, not Slow. Also, I'd suggest using Fast PSY rather than Slow. It's not only faster, it can conserve more bits than Slow PSY. There's a nominal risk of it adversely affecting the quality of your encode, but I've never had a problem with it.

If you're using Standard mode in 5.11, you can still use the MV file, but it may adversely affect the quality. It does still gain some speed, but not quite as much as it used to. I think it skips the MV file for B-frames now, which lessens the performance boost.

If you go with Standard P/Q, B-frames, no MV, and Fast PSY, you should get fairly comparable speed to 5.05. Do give Slowest P/Q a chance with some really low-bitrate encodes though, I've gotten excellent results with just 3 passes, but I usually run 4 just for good measure since I've got a separate box to run the encodes on. Or do what DigitAl56K's been recommending. Run all passes except the last one on Standard P/Q, then run the last one on Slowest. According to him, the difference in quality between that and running all passes on Slowest is almost nil, and saves a lot of time.

[DigitAl56K]

Yes, echoing what others have said:

* Under 5.1.1 "Standard" p/q mode is functionally equivelent to "Slowest" mode under 5.0.x. Using "Standard" mode you should see performance on par with, if not better than, 5.05

* Don't use the MV file, its influence is minimal under 5.1.1 on anything but a legacy machine and it will reduce the PSNR

* Don't use Psycho-vis when doing PSNR tests. Psycho-vis changes the texture in a way that improves perceptual quality for the majority of sources (especially at lower bitrates) but actually degrades the PSNR (changes in the texture equate to noise in this equation). This is a good example of how PSNR is not directly tied to quality - although PV appears to improve quality the PSNR score drops. It also shows that PSNR is really an inadequate method of comparing DivX/XviD etc. because without a subjective test you can't really compare both codecs with their best features enabled.

* Fast Psy is normally a better choice than slow psy. This goes against the common belief that slower=better.

* As Nazgul notes, you can get a big decrease in encoding time if you do all but the last pass in Standard mode. So, for two pass you'd just do Standard, [Slow/Slowest]. This can shave somewhere around 50% off the total time (depending on your system and setup) and during tests the PSNR difference was usually no more than 0.3db and over a number of clips averaged to virtually 0db, indicating that its both safe to use this method and on average no better or worse than doing both passes Slow/Slowest.

Hope this helps.

[SiXXGuNNZ]

I dunno, in standard mode I went from 22 to 18 with 5.11, still alot faster then the 7 fps I get from xvid

[Neo Neko]

*kof* Guys, when you are comparing quality should you not give a codec every chance? With that in mind where was the flaw with using Divx in it's slowest mode? Slowest should give the best quality even if it is only marginally better. I see no flaw. Heck if he used standard who knows. Divx could have scored even worse. There has been work done on Divx, but nothing spectacular. To be honest folks I have not had Divx installed on any of my systems for several months now because of it's lackluster performance. When the next version comes out I will give it a shot of course. But Divx needs to get in gear because while Xvid has flown past it there are even other legal MPEG4 implementations creeping up behind Divx threatening it as the best "Legal" MPEG4 codec. Shitowax and the rest of the 3ivx crew have done some spectacular work lately. If this continues Divx could be in serious trouble real soon. Divx may hold the same relevance as Apple Quicktime MPEG4 real soon. And that's bad.

[BiaTch 5.0]

I vote for a 4 pass encode, 3 standard (I get around 25 fps using -B frames!, XP1700+) & only the last in slowest as recommended by the DivX homepage.

This would come close to using XviD with VHQ4 (speed)

[Zep]

Originally posted by Neo Neko

yup
Originally posted by Neo Neko
Divx may hold the same relevance as Apple Quicktime MPEG4 real soon. And that's bad.

There is no such thing as Apple QT MPEG4 per se'. Apple/qt
uses ISO MPEG4 for encoding and Apple/qt makes standard .mp4
files just like mp4creator.

Apple has to pay fees just like everyone else to use MPEG4/mp4

i think you are thinking that because .mp4 is based on .mov which
the Moving Picture Experts Group (MPEG)and the working group
within the International Organization for Standardization (ISO)
used .mov as the foundation for it. They added to it and tweaked it
and improved it and it is no longer an apple thing.

IMHO the windows side just needs more mp4 tools and players etc...
and why .avi still rules. Well that and for example VD/VDM/Gnot
and on and on all really are geared to .avi (which is why i use
mp4creator on my PC to take the Xvid encode and put it in an .mp4
and stream it to macs. I leave it in .avi for windows users)

[Neo Neko]

Originally posted by Zep
yup


There is no such thing as Apple QT MPEG4 per se'. Apple/qt
uses ISO MPEG4 for encoding and Apple/qt makes standard .mp4
files just like mp4creator.

This I know. When I say Apple Quicktime MPEG4 I am reffering to the MPEG4 codec bundled with Quicktime. Not some proprietary work like MSMPEG4.

Originally posted by Zep
i think you are thinking that because .mp4 is based on .mov which
the Moving Picture Experts Group (MPEG)and the working group
within the International Organization for Standardization (ISO)
used .mov as the foundation for it. They added to it and tweaked it
and improved it and it is no longer an apple thing.

Apple is a major contributor/member to the MPEG group and the ISO.

Originally posted by Zep
IMHO the windows side just needs more mp4 tools and players etc...
and why .avi still rules. Well that and for example VD/VDM/Gnot
and on and on all really are geared to .avi (which is why i use
mp4creator on my PC to take the Xvid encode and put it in an .mp4
and stream it to macs. I leave it in .avi for windows users)

You can leave it in AVI for Mac users as well. Just use Videolan or Mplayer.

[Wolfman]

"Shitowax and the rest of the 3ivx crew"
yes they may be doing good work but is it compatible with standalones, which more and more people are getting? And what is the purpose of 3ivx.. divx is commercial, xvid is open-source and 3ivx is ??? faster/better/less cpu intensive??

[DevilsChild]

3ivx is MPEG-4 compatible, but I think it's oriented more towards the MP4 container. Not sure if anyone has played it on standalone hardware devices...

[Neo Neko]

Originally posted by Wolfman
"Shitowax and the rest of the 3ivx crew"
yes they may be doing good work but is it compatible with standalones,

You should only have to change the fourcc. But the short answer is yes.

Originally posted by Wolfman
which more and more people are getting? And what is the purpose of 3ivx.. divx is commercial, xvid is open-source and 3ivx is ??? faster/better/less cpu intensive??

Well anyone of the 3ivx crew are welcome to correct me. But just by looking at their site they are a commercial MPEG4 codec in the same vein as Divx only they are more aimed at an end to end MPEG4 solution. MPEG4 video, MPEG4 audio, MPEG4 containers. Also embedded solutions seem to be a focus. Perhaps 3ivx certified set-tops would be better.

[jggimi]

This thread is interesting, but is now off topic for DivX encoding. I will split the OT portions into another thread, and move that new thread into the New Codecs or the New Containers forum. But not until I hear back from those involved in this rather feisty conversation as to their preferences.