CoreCodec/H.264 Codec "CoreAVC"

[eddman]

I'll have to get a more powerful processor and was considering these (mostly because they're the fastest 2 and 4 core CPUs with a 65W TDP):

Intel Core i7 2600S with 4 HT cores at 2.8GHz (http://ark.intel.com/products/52215/Intel-Core-i7-2600S-Processor-%288M-Cache-2_80-GHz%29)
Intel Core i3 2125 with 2 HT cores at 3.3GHz (http://ark.intel.com/products/59080/Intel-Core-i3-2125-Processor-%283M-Cache-3_30-GHz%29)

I was wondering effectively CoreeAVC makes use of multiple cores. Is it better to have four cores or two cores with a higher clock speed?

You might want to take a look at these too:

Core i5 2390T / 2 cores / 2.7 GHz / 35 W TDP (http://ark.intel.com/products/53448)
Core i5 2500T / 4 cores / 2.3 GHz / 45 W TDP (http://ark.intel.com/products/52212)

I'd say a quad-core with lower clock speed is better than a dual-core with higher clock speed.

[Gav80K]

You might want to take a look at these too:

Core i5 2390T / 2 cores / 2.7 GHz / 35 W TDP (http://ark.intel.com/products/53448)
Core i5 2500T / 4 cores / 2.3 GHz / 45 W TDP (http://ark.intel.com/products/52212)

I'd say a quad-core with lower clock speed is better than a dual-core with higher clock speed.

Thanks for the suggestions, but I'll probably go with the faster Core i7 2600S to try and keep the skip time as short as possible.

I use the Coolermaster Hyper Z600 which can handle up to an 89W CPU with passive cooling so I should be fine with the 65W i7 2600S.

[nm]

Thanks for the suggestions, but I'll probably go with the faster Core i7 2600S to try and keep the skip time as short as possible.

Well, if skipping was instantaneous with CoreAVC 2.5, it started decoding from the nearest or next keyframe instead of the exact requested frame. If your CoreAVC 3.0 setup does skip to non-keyframe positions and therefore does unnecessary work, I'd suggest finding out the source of the problem instead of throwing hardware and money to work around the issue.

[Gav80K]

Well, if skipping was instantaneous with CoreAVC 2.5, it started decoding from the nearest or next keyframe instead of the exact requested frame. If your CoreAVC 3.0 setup does skip to non-keyframe positions and therefore does unnecessary work, I'd suggest finding out the source of the problem instead of throwing hardware and money to work around the issue.

When I said "pretty much instantly" I meant "an insignificant amount of time that caused no distraction". It now takes about four times as long so you're left with a pause that can last as much as two seconds, which is rather irritating. It does definitely decode from the last keyframe with 2.5.1, otherwise the video would have turned into a mess each time I skipped. It's the same on two PCs with significantly different hardware, so there does seem to be an issue with CUDA acceleration.

With no CUDA support for Hi10P I'd have to get a new PC anyway, so it's not really a major issue, but it is strange the performance has dropped so much.

[ajp_anton]

If you feel adventurous, get a "regular" CPU and undervolt it to get the same result as with a more expensive S or T model. Also, you have access to higher performance when seeking (which is so fast your cooling won't be affected).

[cyberbeing]

We are planning to do a release this week.

Is there a tentative changelog for 3.02 at this point?

[RoadPizza]

Thank you for the answers to my question about software vs hardware decoding. I have one other question that I have seen asked several times in this thread, but never answered. What, exactly, does the deblocking option "skip when safe" do? Thanks!

[BetaBoy]

Is there a tentative changelog for 3.02 at this point?

Its a floating target atm as we want to address both the bugs and speed concerns. But we were side tracked a few days as we have been working on the CoreAVC Developer SDK.... which is now updated to 3.0.2.

[nibus]

Does the new CoreAVC 3 decode 10-bit without the change in colors that is/was present in ffdshow? And has anyone really noticed big improvements from 10-bit in general?

[squid_80]

What, exactly, does the deblocking option "skip when safe" do? Thanks!

From the Help page in the filter properties:
Skip when safe - skip deblocking step when decoding B-frames.

[RoadPizza]

From the Help page in the filter properties:

I found that of course. I know I'm a newbie, but give me a little credit for having searched for my answer first :-)

I'm looking for a bit more detail... For example, from post #3890 in this thread:

"Do you mean NAL units designated as disposable (unreferenced B-frames), or do you really mean all B-frames? If the latter, its certainly not safe."

Also, if it really is safe to do this, why isn't it just part of the spec in the first place? Or at least the CoreAVC default? Thanks for any info.

[squid_80]

It means unreferenced B-frames. Skipping deblocking is not part of the spec at all, it degrades the image quality. The term "safe" means the image errors caused by lack of deblocking won't accumulate over time (because they will only apply to individual frames that aren't referenced by others).

[flapane]

With latest version of CoreAVC (2.5.5) on 7 x64 and HD5670, DXVA worked great with MPC-HC, but I couldn't enable it on WinTV live HDTV. The tray icon remains blue and the cpu sits at about 60%.
Furthermore, a guy from the WinTV tech support asked for more informations on CoreAVC API needed to set the GPU acceleration. I suppose that it may help to better understand the problem.
Any hints?
Thanks

Tried 3.0.1, nothing changed. The tray icon is still blue.

[nussman]

Take a look at page 331.
The infos für WinTV tech support could be found there!

[CiNcH]

Guess this is the same problem as with the DVBViewer. CoreAVC requires the splitter to deliver SPS/PPS at connection time. If the splitter does not deliver, CoreAVC falls back to SW mode.

[nussman]

@flapane: Please let us know how WinTV tech support thinks about it!

[CiNcH]

@flapane: Please let us know how WinTV tech support thinks about it!
They won't have any opinion. Believe me ;) . Does WinTV use the MS MPEG-2 demultiplexer?

[flapane]

I think they'd gladly work together with CoreAVC devs in order to solve the issue, otherwise we'll never get HW acceleration: http://www.hauppauge.co.uk/board/showthread.php?22237-CoreAVC-DXVA-support&p=103351#post103351

[CiNcH]

I now tried WatchTVPro. It uses the MPEG-2 Demultiplexer. Most DVB BDA apps do. The result is no DXVA (see here (http://members.inode.at/762450/coreavc/mpeg_demux_coreavc.jpg)). The WinTV guys can't solve this issue. Only if they develop their own splitter which adheres to the CoreAVC rules.

[flapane]

As you can see on Hauppage forum, they sent an e-mail to CoreAVC devs, without luck.

[CiNcH]

CoreCodec is aware of the issue. All you can do is sit back an wait (hope). nussman posted a reference to the discussion with technical background...

[dead_screem]

betaboy, any news on 3.0.2?

[D.A.S.]

1. Not working DXVA
2. OS - Windows 7 SP1 32bit
3. CPU - AMD Phenom II X4 940
4. GPU - Radeon HD 5670
5. Video Driver version - 11.8
6. Player - MPC-HC v1.5.3.3739
7. Renderer - EVR custom pres **
8. Splitter - Haali 1.11.96.14
9. Codec - CoreAVC 3.0.1
10. Output - NV12
11. Acceleration - DXVA on.

Sample
http://www.multiupload.com/4RR117CUFZ

Built-in decoder MPC-HC works with DXVA.

If the file is repackage a new version of mkvmerge then DXVA to work on CoreAVC 3.0.1

[kirakami]

when will CoreAVC 3.0.2 release?
will CoreAVC 3.0.1 also work on my old pc's with Nvidia Ge-Force 4 MX440?
what are steps to enable HW acceleration on CoreAVC with MPC-HC?

[BetaBoy]

An update for all... we are taking a little more time in devel and going right to CoreAVC 3.1 and adding more features, speed enhancements then planned. On apps that that do not conform to proper SPS handling... we added a work around for 3.1.

This new work might delay adding 4:4:4 profile support (which is mostly done anyway) but based on the feedback we feel the potential delay is worth it.

[Mangix]

will CoreAVC 3.0.1 also work on my old pc's with Nvidia Ge-Force 4 MX440?

no. cpu decoding still works though.

that card was released in 2003 i think. 8 years wow...

[mkanet]

I just upgraded my display card to a 4th generation Nvidia Purevideo HD card.. GT545. I was told it has the best decoding that Nvidia has to offer.

I am curious which of the following would take advantage of my graphics card's decoding capabilities the best:

1. Windows 7 64bit "DTV-DVD Video Decoder"
2. CoreAVC 3.0.1 Cuda
3. CoreAVC 3.0.1 DXVA


Above, when I mean "best", I mean in respect to bluray spec H.264 decoding.

Shamelessly, I really, REALLY wanted LAV Video decoder to work since it also supports mpeg2/VC1; however, on my machine, it has unacceptable stability issues under my directshow software. CoreAVC, "knock on wood" works no matter settings I choose... just not sure which is ideal for my needs.

[Asmodian]

Well the GT 545 (I have one too) isn't actually any better than any of the rest of the fermi cards, only the GT520 (as far as I know) has Purevideo HD feature set D, the fifth generation of purevideo.

That said, the GT 545 is one of the better cards as the GT520 doesn't have enough cores to do the best deinterlacing.

CoreAVC Cuda is what you want.

[mkanet]

Thanks. I mean the GT 545 was better than generation 3, 2, and 1. I upgraded from an GT8500.

What's the difference between CoreAVC Cuda and the decoder that comes with Win7?

If I want cuda for mpeg2 and VC1, what decoder can I use if LAV Video has stability issues on my system?

EDIT: I just realized, I also have latest version of TMT5 installed on my machine. I already use it's audio decoder. Doesn't it's video decoder already support cuda for all popular video formats? I vaguely remember that TMT started supporting cuda back in version 3.

This may sound like a dumb question, but isn't this the same kind of cuda that's used in CoreAVC and LAV Video?

Well the GT 545 (I have one too) isn't actually any better than any of the rest of the fermi cards, only the GT520 (as far as I know) has Purevideo HD feature set D, the fifth generation of purevideo.

That said, the GT 545 is one of the better cards as the GT520 doesn't have enough cores to do the best deinterlacing.

CoreAVC Cuda is what you want.

[Mixer73]

This may sound like a dumb question, but isn't this the same kind of cuda that's used in CoreAVC and LAV Video?

You're getting confused by the fact that there's a few different paths to the same hardware decoders.

CoreAVC and TMT will use standard DXVA, where you just offload the processing to the hardware and it gets played.

CUDA is a programming interface to use the shaders on your graphics card, but it also allows a function called CUVID which is a way to interface to the DXVA decoding and let you have access to capabilities which are not available through standard DXVA - like subtitles, madVR etc. I think CoreAVC labelling this "CUDA" is a bit of a misnomer.

Now I'm not half as smart as a lot of the guys here but I hope this clears things up for you.

[LigH]

It has been explained and criticized dozens of times already. The hardware accellerated video decoders don't use CUDA to calculate the decoding, they just use it to let the video pass to the PureVideo hardware decoding engine (V2 or newer). "Misnomer" is indeed a matching rating ... "CUDA" sounds so conveniently advertizing; "CUVID" instead is less famous, although more correct.

[SEt]

It does use CUDA, but not for decoding h264/vc1 itself - for reading decoded data from videocard. It's quite expensive process too.

mkanet, it was said many times that all decoders that correctly decode h264 are equal. Zero quality difference, be it hardware or software. Nowadays there is no much reason for caring about hardware decoding (unless your CPU is seriously outdated or you want to save every possible bit of energy or want hardware deinterlacer). As about problems with LAV - they are likely due to something bad installed on your system (hint: codecpacks).

Also even if different codecs use the same DXVA or CUDA decoders - it doesn't mean they work exactly the same. Because there are plenty of places before and after them where mistakes can be made.

[CruNcher]

It has been explained and criticized dozens of times already. The hardware accellerated video decoders don't use CUDA to calculate the decoding, they just use it to let the video pass to the PureVideo hardware decoding engine (V2 or newer). "Misnomer" is indeed a matching rating ... "CUDA" sounds so conveniently advertizing; "CUVID" instead is less famous, although more correct.

Actually for Mpeg-2 Bitstreams it does use CUDA to fix a Hardware issue in VP2 upto VP4 ;)

Also even if different codecs use the same DXVA or CUDA decoders - it doesn't mean they work exactly the same. Because there are plenty of places before and after them where mistakes can be made.

Yep true true even DXVA decoder can differ greatly especially in efficiency and bugs i remember the 60 fps playback problem on my VP2 that only Cyberlink was able to cope with on VMR9 compared to all the other DXVA implementations (which was really unique) :P
And imho it is still the most efficient DXVA implementation for both NT 5/6 and GPUs compared to others, though it also has it's drawbacks with some not so widespread bitstreams that other implementation again do better with :)

[jmelan]

any update on a release date for coreMVC?

[cyberbeing]

Progress report on CoreAVC 3.1 as well?

[CiNcH]

"QA" must be hard at work...

[BetaBoy]

We finished our current todo for our 3.1 core work. We have added 4:4:4 profile support and fixed most of the reported bugs as well as added a bunch of 10 bit optimizations.

We still have some directshow work to do, so let's see how the next week work goes.

[BetaBoy]

any update on a release date for coreMVC?

With the recent microsoft additions to MVC in directshow/windows 7. We will release a consumer decoder after we add those additions. Also we wanted to wait till we added 4:4:4 support which is now done. So likely sometime at or around CES.

Note that the CoreMVC Library SDK (non-directshow) has aleady been released for Developer / OEM integration and is being used for consumer 3D applications.

[robpdotcom]

Any news on getting a fix for the Haali/TrueHD issue?

[Livias]

@BetaBoy, are you talking about CES as in next January?

[Mitchan81]

I recently purchased CoreAVC 3.0.1 for use DXVA with my Intel G45 GPU - X4500HD

Now I was looking to buy an NVIDIA card (Intel have 23.976hz playback issue) but I did not understand about what parameters should I look to use all features of CoreAVC's CUDA decode? ram? gpu clock? number of cuda core? VPA3-4-5? etc etc

which card should I buy?

[LoRd_MuldeR]

I recently purchased CoreAVC 3.0.1 for use DXVA with my Intel G45 GPU - X4500HD

Now I was looking to buy an NVIDIA card (Intel have 23.976hz playback issue) but I did not understand about what parameters should I look to use all features of CoreAVC's CUDA decode? ram? gpu clock? number of cuda core? VPA3-4-5? etc etc

which card should I buy?

Shouldn't matter, because DXVA as well as "CUDA acceleration" (actually CUVID) doesn't use the GPU at all!

DXVA and CUVID are just two different API's to access the card's built-in "video engine" (called "PureVideo" in NVidia jargon). But these are separate/dedicated circuits, not the actual GPU!

And, most important, the video engine is pretty much identical between "low end", "performance" and "high end" cards.

(Just be sure you buy a card that has the latest incarnation of the "PureVideo" engine on board. The latest version currently is VP5. Funny enough, the "high end" cards still use VP4)

[Mitchan81]

(Just be sure you buy a card that has the latest incarnation of the "PureVideo" engine on board. The latest version currently is VP5. Funny enough, the "high end" cards still use VP4)

Exact!!! The only one with VP5 is GT520 (64bit bus, 1gb ddr3)
So... you're telling me that gt520 is faster then gtx590 (384bit x 2 bus, 3072 MB ram etc etc) with CorAVC - CUDA acceleration (PureVideo HD) ?

[nm]

GT 520 decodes faster (H.264 video at 120 fps instead of 60 fps), but if you also want to postprocess (deinterlace, denoise, ...) HD video on the GPU, then the other parameters such as the number of CUDA cores do matter. GT 520 may not be fast enough when using multiple filters at the same time -- it can barely manage the "temporal-spatial" deinterlacing mode for 1080i30.

[nevcairiel]

I would not buy the 520, its too slow. Go with a 430 or 440, or if you want some more, a 450 or 550.

[Mitchan81]

I would not buy the 520, its too slow.

slow for what? LAV CUVID or madVR?


if you want some more

yes.. but more WHAT??? :D

I need only HTPC video card. No games.

thx for your support ;)

[nevcairiel]

The 520 is too slow for anything. It can't even properly do full quality deinterlacing, be it with EVR or anything else.

[eddman]

I recently purchased CoreAVC 3.0.1 for use DXVA with my Intel G45 GPU - X4500HD

Now I was looking to buy an NVIDIA card (Intel have 23.976hz playback issue) but I did not understand about what parameters should I look to use all features of CoreAVC's CUDA decode? ram? gpu clock? number of cuda core? VPA3-4-5? etc etc

which card should I buy?

Here, read up and decide.

http://www.anandtech.com/show/4380/discrete-htpc-gpus-shootout/1

[rica]

Hi guys!

I just purchased CoreAVC to test its mvc deoding capability.
Can somebody guide me how it is possible on GraphStudio?

Which splitter, which renderer?

Thanks.


_ _ _ _

[LoRd_MuldeR]

I just purchased CoreAVC to test its mvc deoding capability.

Aren't CoreAVC and CoreMVC two different products, with the latter still to be released (as a standalone product)? :confused:

http://forum.doom9.org/showpost.php?p=1535817&postcount=6889