As a Decoder, CoreAVC or DivX 7?

[xLusi0n]

So, I'm looking for a H.264 decoder and was about to get the CoreAVC one...BUT DivX 7 is out now and also has H.264 decoder...after reading the long DivX 7 thread...I don't know which to get.

I believe the decoding speeds and system usage is similar?

Does DivX 7 only decode H.264 in MKV? I know it only encodes to MKV, but does it decode in other formats?

Thanks.

[Atak_Snajpera]

DivX 7 is for free and offers very good decoding speed (fully multi-threaded) . CoreAVC is probably faster but it costs $. You should also try FFDshow MT.

Small test on my Q6600@3Ghz
Source is HF100 comcorder 16Mbps 1920x1080i50
Results: FFDshow MT -> 186 dfps
Divx7 Decoder -> 225 dfps
CoreAVC -> 216 dfps

[Cyber-Mav]

divx7 for me seems to be a bit quicker on a single core opteron 146 2ghz cpu in decoding compared to coreavc 1.85.

[tetsuo55]

divx7 for me seems to be a bit quicker on a single core opteron 146 2ghz cpu in decoding compared to coreavc 1.85.

Intersting.

I would love to see a comparison between the latest builds of divx7, coreavc and ffdshow on a single core CPU.

Preferably one without SSE2

[IgorC]

It's possible to download trial version of CoreAVC and compare it with Divx 7.

Benchmark for comparison of decoders perfomance http://haali.cs.msu.ru/mkv/timeCodec.exe

Both decoders should be on par. CoreAVC should be a bit faster on older non-SSE2 CPU while Divx on newer CPU with SSE2 support. But in the last case the difference won't be important.

[xLusi0n]

Interesting...I'll probably go with DivX and pay 20 bucks to have encoding too. Thanks!

[Atak_Snajpera]

You are going to pay 20$ for encoder which is alot worse than x264?!?!?!

[Jay Bee]

AFAIK divx doesn't support HW deinterlacing (or any deinterlacing?).

[Sagekilla]

If you want HW deinterlacing of any sort when encoding, only way (I know of) is through neuron2's DGNV*Dec. I don't know of any other kinds available right now.

[Kurth]

AMD Athlon 64 X2 4000+ 2.1 Ghz
Using timecodec.exe with a video encoded with H264 resolution 1920x1080
ffdshow dfps: 34.1
ffdshow MT dfps: 53.1
Divx 7 dfps: 83.5
CoreAVC 1.8.5 dfps: 101.6

Nice result on the Divx decoder but CoreAVC still better than Divx on my Dual Core processor.

[Jay Bee]

If you want HW deinterlacing of any sort when encoding, only way (I know of) is through neuron2's DGNV*Dec. I don't know of any other kinds available right now.

But the OP needs a decoder.

[DigitAl56K]

In comparison to CoreAVC, the DirectShow version is currently missing real-time deinterlacing but you may not need this and there is already a new build in the works, keep an eye on the Project Rémoulade (http://labs.divx.com/projectremoulade) group at DivX Labs for updates. DivX Player 7 on the other hand will do de-interlacing during playback already.

[philipshu]

Core 2 Duo T7500 deblocking off, VMR 9
CoreAVC 1.8.5: 76.5 dfps
DivX 7: 81.1 dfps
looks like DivX is free and faster

[Astrophizz]

Why'd you turn deblocking off?

[littleD]

Intersting.

I would love to see a comparison between the latest builds of divx7, coreavc and ffdshow on a single core CPU.

Preferably one without SSE2

My results with Sempron@1.8 GHz, no SSE2
Short sample 1920x800 11Mbps, from bluray
All values are dfps@VMR9

FFDshow: 6,5
skip deblocking when safe: 6,7
CoreAVC: 18,9
skip deblocking when safe: 19,6
Divx H264: 17,2

At this point of development i consider CoreAVC better as it's faster and has more useful options, but i think divx can be still improved in near future.

[Atak_Snajpera]

http://forum.doom9.org/showthread.php?p=1236822#post1236822

I've added results for CoreAVC.

[ajp_anton]

Core 2 Duo T7500 deblocking off, VMR 9
CoreAVC 1.8.5: 76.5 dfps
DivX 7: 81.1 dfps
looks like DivX is free and fasterI've been told (plus common sense) that turning off deblocking is one of the stupidest things to do if the CPU isn't too slow.

[Atak_Snajpera]

@DigitAl56K
At the moment current version can not detect fps value when used in AviSynth script.

http://img396.imageshack.us/img396/576/snapshot20090114112144ed5.th.png (http://img396.imageshack.us/my.php?image=snapshot20090114112144ed5.png)

I have no problems with FFDShow or CoreAVC. It happens only with Divx decoder.

[Jay Bee]

In comparison to CoreAVC, the DirectShow version is currently missing real-time deinterlacing but you may not need this and there is already a new build in the works...

Could be interesting for DVBViewer users as up to now Cyberlink is about the only decoder out there that doesn't suffer from small stutters on 1080i live TV.

[BetaBoy]

I've been told (plus common sense) that turning off deblocking is one of the stupidest things to do if the CPU isn't too slow.

I love it when ppl like philipshu interject their newb wisdom.

[Betsy25]

Please excuse me for being noob.

I don't understand this whole fps comparison, I mean all movies are practically encoded at 29.76fps, so why this comparing decoders that can do 100+ fps, any decoder capable of say 30fps should be enough ? Or am I wrong ?

[Jeff Flowerday]

Please excuse me for being noob.

I don't understand this whole fps comparison, I mean all movies are practically encoded at 29.76fps, so why this comparing decoders that can do 100+ fps, any decoder capable of say 30fps should be enough ? Or am I wrong ?

It's a meter for comparison nothing more. Looking at the results the one that pushes more frames per second if more efficient and will run smoother on lower end hardware.

[Dark Shikari]

Please excuse me for being noob.

I don't understand this whole fps comparison, I mean all movies are practically encoded at 29.76fps, so why this comparing decoders that can do 100+ fps, any decoder capable of say 30fps should be enough ? Or am I wrong ?1. Slower CPUs, obviously, will decode slower, so 50fps on a fast CPU might mean 25fps on a slower one, at which point it really does matter.

2. The less CPU a decoder uses, the better. For example, I run a bunch of postprocessing filters on Starcraft VODs because of the relatively low quality. If the decoding took up too much CPU time, I wouldn't have any left for the filters.

3. If a decoder can do 60fps at 720p, it probably can only do 25fps at 1080p.

[Betsy25]

Thanks Jeff & Dark Shikari, I didn't think about it that way.:cool:

[xLusi0n]

So, I download both and tried both CoreAVC and DivX 7 to see which one was better as a decoder.

On a Intel Centrino Duo 2.00 laptop with 2GB RAM, I got the following results

CoreAVC:
User: 8s, Kernel: 0s, Total: 9s, Real: 35s, FPS: 380.2, DFPS: 92.1

DivX 7:
User: 9s, Kernel: 0s, Total: 10s, Real: 31s, FPS: 316.4, DFPS: 101.8

I don't know what all the numbers mean, but I'm guessing DFPS is what I care about?

All the settings were default, I didn't make any tweaks to DivX 7 or CoreAVC.

[leeperry]

keep an eye on the Project Rémoulade (http://labs.divx.com/projectremoulade) group at DivX Labs for updates.
if this time you could send emails to betatesters when updates are out, that'd be much appreciated :D

[mariush]

None.

Just start Media Player Classic Homecinema and my ATI 4850 will take care of everything in hardware.

[Nazgulled]

I download 2 trailers from DivX showcase page and did some testing between CoreAVC vs DivX 7 H.264:

===========================================================================================
DECODER: CoreAVC
===========================================================================================
STAR TREK TRAILER (720p)
-------------------------------------------------------------------------------------------
- User: 2s, kernel: 0s, total: 2s, real: 19s, fps: 1390.4, dfps: 158.0
- User: 1s, kernel: 0s, total: 1s, real: 21s, fps: 1575.0, dfps: 146.0
- User: 2s, kernel: 0s, total: 2s, real: 20s, fps: 1550.8, dfps: 157.0
- User: 1s, kernel: 0s, total: 2s, real: 20s, fps: 1562.8, dfps: 156.6
- User: 2s, kernel: 0s, total: 2s, real: 19s, fps: 1493.3, dfps: 159.0
-------------------------------------------------------------------------------------------
AVERAGE: 155.0
===========================================================================================
WATCHMEN TRAILER (1080p)
-------------------------------------------------------------------------------------------
- User: 5s, kernel: 0s, total: 5s, real: 45s, fps: 670.7, dfps: 75.2
- User: 4s, kernel: 0s, total: 4s, real: 49s, fps: 733.7, dfps: 68.5
- User: 4s, kernel: 0s, total: 4s, real: 44s, fps: 751.4, dfps: 75.9
- User: 4s, kernel: 0s, total: 4s, real: 45s, fps: 691.9, dfps: 74.4
- User: 5s, kernel: 0s, total: 5s, real: 45s, fps: 652.7, dfps: 75.5
-------------------------------------------------------------------------------------------
AVERAGE: 73.9


===========================================================================================
DECODER: DivX 7
===========================================================================================
STAR TREK TRAILER (720p)
-------------------------------------------------------------------------------------------
- User: 3s, kernel: 0s, total: 3s, real: 22s, fps: 900.0, dfps: 141.3
- User: 3s, kernel: 0s, total: 3s, real: 21s, fps: 969.2, dfps: 143.3
- User: 3s, kernel: 0s, total: 4s, real: 21s, fps: 778.4, dfps: 144.7
- User: 2s, kernel: 0s, total: 2s, real: 21s, fps: 1244.5, dfps: 146.8
- User: 2s, kernel: 0s, total: 2s, real: 21s, fps: 1185.9, dfps: 146.6
-------------------------------------------------------------------------------------------
AVERAGE: 144.5
===========================================================================================
WATCHMEN TRAILER (1080p)
-------------------------------------------------------------------------------------------
- User: 5s, kernel: 0s, total: 5s, real: 48s, fps: 570.9, dfps: 70.8
- User: 6s, kernel: 0s, total: 6s, real: 53s, fps: 524.3, dfps: 63.6
- User: 6s, kernel: 0s, total: 6s, real: 48s, fps: 519.4, dfps: 70.3
- User: 6s, kernel: 0s, total: 6s, real: 48s, fps: 513.3, dfps: 70.9
- User: 5s, kernel: 0s, total: 5s, real: 48s, fps: 605.7, dfps: 70.4
-------------------------------------------------------------------------------------------
AVERAGE: 69.2

I don't know what all this means but I suppose that, at least on my computer, CoreAVC is faster/better?

Now, the question is: does the difference between them is worth buying CoreAVC or should I go the DivX 7 route?

[Nazgulled]

I think they are both good in quality. But this is just an opinion I haven't tested anything in regards to picture quality.

Like I said before, I don't know what all those numbers mean, but all posts seem to mention that we should take a look at the dfps value and the higher the better. However, I just used Process Explorer from Microsoft, which allows me to show a CPU usage graph individually by process and the process from MediaPortal (it's the software I use for MediaCenter) was lower compared to CoreAVC while playing 1080p content.

I guess I'll just go with DivX which:
- It seems to use less CPU
- It's free...
- Doesn't have the annoying bug that CoreAVC has that popups the taskbar while watching in full screen

[Dark Shikari]

I tried both Core AVC Pro & D.264.
Deblocking quality of Core AVC Pro was horrible, D.264 is much better than Core AVC Pro in quality, I think.If two H.264 decoders differ in quality, you're doing something wrong. H.264 decoders are required to give bit-identical output.

[LoRd_MuldeR]

If two H.264 decoders differ in quality, you're doing something wrong. H.264 decoders are required to give bit-identical output.

I think there are two common problems that can cause different output:
* Deblocking: Some decoders (optionally) skip deblocking for speed-up, which leads to "corrupted" output, of course.
* Different (wrong) luminance levels (http://forum.doom9.org/showthread.php?t=143689)

[Dark Eiri]

None.

Just start Media Player Classic Homecinema and my ATI 4850 will take care of everything in hardware.

Yeah, right. Unless your video is not DXVA compliant.

I would stick with DivX 7. If your CPU is fast enough, it will make no difference.

[lucassp]

I would stick with DivX 7. If your CPU is fast enough, it will make no difference.

Yes it will. The GPU can do lots of things like: denoise, deinterlace and pulldown, without putting any load on your CPU.

http://www.xbitlabs.com/articles/video/display/media-playback_6.html

[Dark Eiri]

^ Yes, I know. I replied to the topic owner, between CoreAVC and DivX 7, I would stick to DivX 7. He didn't mention GPU decoding, and I doubt he would want to buy another GPU just to decode H264 and still be stuck with DXVA compliant videos only.

[BetaBoy]

Why do I feel that ppl are ignoring Dark Shikari's input. ;-/

Sina... you are the first person to report such an issue. Do you have the source material we can take a look at?

[LoRd_MuldeR]

what do you mean source material?

Post the video file that can be used reproduce the problem you described.

(They can't fix it unless they are able to reproduce it)

[tommy_vercetti]

I asked this somehere else, but do you guys notice any audio delay when playing a .MP4 file using DivX 7 with Haali splitter

Im on Vista x64, 4 GB RAM, Movie : Lord of Rings - Twin Towers, ripped to x264 using FairUse, the movie plays fine with VLC or FFDshow

The Video rendering is ok, no issue regarding that, did not try to check it from Graphedit

[HowlerX]

I downloaded the video Sina© posted and aside from the video being pretty much unwatchable due to its low bitrate, I noticed no "lag" with CoreAVC or DivX h.264 decoders.

[DJ Bobo]

As far as I'm concerned, I only believe in one index: CPU load!
On my laptop, equipped with an AMD Athlon X2 QL-60 (1.9GHz), the figures are as follows:

CoreAVC 1.8.5: 72% avg (84% max)
DivX 7: 84% avg (94% max)

So the winner is clearly CoreAVC for me.
Test was done on the same clip (1920x800@24fps) in MPC HC using the system monitor of Windows Vista.

Hope this helps.

[CruNcher]

imho haalis timecodec can be sometimes giving out confusing results with certain decoders (as it's bound to haalis splitter and some decoder dont like that, others wont even connect) i changed my testing setup therefore for absolute speed and rendering time to graphstudio, it does a 3 (you can define that yourself) run check and the results are very accurate and also dont seem to fluctuate that much, you also realize very fast wich Renderer are refresh caped with it :)

[allouh]

gonna post this somewhere else.

[dvy]

As far as I'm concerned, I only believe in one index: CPU load!
On my laptop, equipped with an AMD Athlon X2 QL-60 (1.9GHz), the figures are as follows:

CoreAVC 1.8.5: 72% avg (84% max)
DivX 7: 84% avg (94% max)

So the winner is clearly CoreAVC for me.
Test was done on the same clip (1920x800@24fps) in MPC HC using the system monitor of Windows Vista.

Hope this helps.


CoreAVC indeed faster than DivX 7!!!!

[davidmoore]

Is Divx7 64bit?

Edit: NM, it's not.

[davidmoore]

C2D E6600 @ 2.4ghz
4GB of DDR2 1066
8800 GTS 640mb
Renderer: VMR9
720p Death Note Special [Kuro-Hana]
FFDShow tryouts 2844: DFPS: 50.6 CPU%: ~70%
Divx 7: DFPS: 102.8 CPU%: ~95%
CoreAVC 1.9.5 w/ CUDA: DFPS: 119.3 CPU%: ~60%

Keep in mind that the 8 series Nvidia cards barely utilize CUDA.

The CPU% are guesstimates based on the CPU utilization during the tests and my perception of the averages.

[nm]

Keep in mind that the 8 series Nvidia cards barely utilize CUDA.
You mean G80 cards. For example, 8800 GTS 512/1024 MB (G92) has VP2, which CoreAVC can use through CUDA & nvcuvid.dll. Your card has VP1, so CoreAVC doesn't use CUDA at all. VP1 can accelerate H.264 decoding through DXVA, but it isn't very helpful (and deblocking may be broken with some decoders).

[davidmoore]

You mean G80 cards. For example, 8800 GTS 512/1024 MB (G92) has VP2, which CoreAVC can use through CUDA & nvcuvid.dll. Your card has VP1, so CoreAVC doesn't use CUDA at all. VP1 can accelerate H.264 decoding through DXVA, but it isn't very helpful (and deblocking may be broken with some decoders).

I knew there was a difference between the cards which is why I put 640mb. But I wasn't sure about the G80/G92. I also wasn't aware that my card has zero CUDA support. We learn new things everyday. I guess that being said, CoreAVC w/ CUDA would be even better but also makes the benchmarks I posted less biased.

[nm]

I also wasn't aware that my card has zero CUDA support.
To be precise, it supports CUDA otherwise, but not video decoding with NVCUVID which NVIDIA included as an extension to CUDA 2.0. CoreAVC's CUDA support is based on this video decoding API.

I guess that being said, CoreAVC w/ CUDA would be even better but also makes the benchmarks I posted less biased.
Yep, although the CPU usage that you observed (60% for CoreAVC) is strange. I'd expect 90-100% with software decoding and 1-20% with hardware decoding.

[davidmoore]

To be precise, it supports CUDA otherwise, but not video decoding with NVCUVID which NVIDIA included as an extension to CUDA 2.0. CoreAVC's CUDA support is based on this video decoding API.


Yep, although the CPU usage that you observed (60% for CoreAVC) is strange. I'd expect 90-100% with software decoding and 1-20% with hardware decoding.

Should I rerun the test again with CoreAVC w/ CUDA support disabled (see if it makes a difference)? I also had hardware deinterlacing on.

Edit: I just unchecked Prefer CUDA and put Deinterlacing on Single Field.

Edit2: It looks like I'm using around 50%-55% CPU now... errr...

[rahzel]

My results with an AMD BE-2350 X2 @ 2.3Ghz.
ffdshow MT - 21.6dfps
Divx 7 H264 - 35.6dfps
CoreAVC - 36.9dfps

I have a 1080p video that is very challenging to play without stuttering as it stresses the CPU a lot, and CoreAVC does a noticeably better job with it. My CPU utilization is also consistent with the timecodec results. With Divx and CoreAVC playing a certain 1080p video, my CPU utilization is anywhere between 40-60%, but with ffdshow MT, it's usually between 55-80%.

For a free decoder, I find Divx to pretty darn good. I noticed playing certain 1280x528 videos with CoreAVC, there are small black bars on the sides when put in full screen mode (MPC HC, stretch to inside, keep AR checked) whereas most 1280x(xxx) videos won't have these bars. Using Divx H264, MPC HC's built in decoder, or FFDshow, these bars don't show up. So I think I'm actually going to use Divx as the difference between the two seem to be negligible.

One question though -- is there any way to set the output level (TV 16-235, or PC 0-255) with Divx 7 like you can with CoreAVC? With my ATI HD3200 connected to my Samsung LCD with HDMI and using YCbCr422 in ATI CCC, the levels appear correct, but just wondering as I build many HTPC's for people, and I may just use this instead.

[jwdaigle]

Maybe I'm the only one that thinks this, but for me, its all about picture quality.

I couldnt care less how busy the CPU is, and whether I am using GPU acceleration or not. It just doesnt matter.

Dark Shakiri made a point (which I think is a good one) that all decoders are required to have bit perfect output, which makes them identical.

But aren't there post processing algorithms that some have and some dont? And wouldnt they affect picture quality?

I am currently using CoreAVC 1.9.5, and love it. Im using it with an nVidia 9400 IGP & E7400, and Media Portal.