madVR - high quality video renderer (GPU assisted)

[leeperry]

this could help finding the culprit: http://www.resplendence.com/latencymon

it's almost always due to sloppy drivers hogging the DPC waiting line.

[cyberbeing]

Yeah, for the longest time on my WinXP box I was having horrible DPC latency issues. It turned out to by my SATA drivers. All drivers from late 2006-2008 were borked. Needed to roll back to early 2006 drivers. :rolleyes:

http://www.resplendence.com/latencymon
First time seeing that program *bookmarks*. It seems to fall between Microsoft's xperf and rattv3 functionality wise. All pretty much do the same thing though.

[pankov]

leeperry,
that seems like a very nice tool but it's kind of not consistent with DPC Latency checker. If I start monitoring with LatencyMon without madVR it shows low values (~80us) and then if I start playback with madVR it doesn't jump, while at the same time DPC Lat checker jumps in the ~750 range. If I start monitoring with madVR playing it shows the correct value ~750 and shows that ndis.sys is the one that's taking 600-700us above the normal 80-100us times that I get without madVR.
The strangest thing is that it's only madVR that's affected.
I tested it with 1080p test clip without audio to rule out the sound card and all other renederers don't have this problem (VMR/EVR/Haali).
So it has to be something in the video processing that madVR is doing on the GPU. This made me think that it's Catalyst drivers that I use so I decided to confirm it and ran some 3D test (Furmark, 3D Mark) and games (FIA WRC, F1) and to my surprise none of them showed high latency.
Now I'm totally lost.
I'm happy though that these high (~750us) times don't affect my playback and I don't get any drop-outs or pops in the sound or lags in the video ... yet

[cyberbeing]

You may want to re-confirm that with xperf or rattv3, the Microsoft tools. They shouldn't have the issue you're seeing with latencymon.

If correct, NDIS is this (http://en.wikipedia.org/wiki/Network_Driver_Interface_Specification). Which means something on your system has an issue with madVR's network functionality. Anti-Virus, Anti-Spyware, Firewalls, network monitoring tools, etc, disable them all. Reboot and see if the problem goes away. If not, uninstall the drivers for all Network Adapters completely. Reboot (don't let the network drivers re-install on boot).

[Plutotype]

leeperry,
that seems like a very nice tool but it's kind of not consistent with DPC Latency checker. If I start monitoring with LatencyMon without madVR it shows low values (~80us) and then if I start playback with madVR it doesn't jump, while at the same time DPC Lat checker jumps in the ~900 range. If I start monitoring with madVR playing it shows the correct value ~900 and shows that ndis.sys is the one that's taking 600-800us above the normal 80-100us times that I get without madVR.
The strangest thing is that it's only madVR that's affected.
I tested it with 1080p test clip without audio to rule out the sound card and all other renederers don't have this problem (VMR/EVR/Haali).
So it has to be something in the video processing that madVR is doing on the GPU. This made me think that it's Catalyst drivers that I use so I decided to confirm it and ran some 3D test (Furmark, 3D Mark) and games (FIA WRC, F1) and to my surprise none of them showed high latency.
Now I'm totally lost.
I'm happy though that these high (~900us) times don't affect my playback and I don't get any drop-outs or pops in the sound or lags in the video ... yet

I have restarted the computer and the Latency mon works ( also thought its not working.. )
Now its clear, that it is the RME babyface causing the DPC issues:
http://postimage.org/image/hrmw3xtw/

RME Forum moderator wrote, that I have either a hardware issue in my computer or babyface is not installed properly (adviced to dl latest drivers ). What is then causing these DPC counts 2000-3999?

Parallely RME soundcard gui showed me USB-transmission errors while Latency Mon has recorded the events in the background.
http://postimage.org/image/hurcnjes/

What do you think, its a USB port issue? Or its a soundcard issue?

Thanks
Pluto

[leeperry]

that it is the RME babyface causing the DPC issues
it seemed obvious, but I didn't dare saying it right away.

if everything's fine when it's not in use...try another USB interface to find out who the culprit really is(the RME drivers or your USB adapter drivers).

anyway, it's a bit OT at this point...might wanna save madshi from getting through all this, as his software is not at fault ;)

[Plutotype]

it seemed obvious, but I didn't dare saying it right away.

if everything's fine when it's not in use...try another USB interface to find out who the culprit really is(the RME drivers or your USB adapter drivers).

anyway, it's a bit OT at this point...might wanna save madshi from getting through all this, as his software is not at fault ;)

Hi,
After further testing, huge DPC spikes has been notified by the LatencyMon at following drivers/sys files:

usbport.sys ( or any name assigned by the usb controller )
portcls.sys
ataport.sys
scsiport.sys

I uninstalled the RME, same latency results came also with onboard Realtek. I did use this chip before, but never noticed any sound issues. The issue seems related more and more to my hardware. If somebody has X58a UD3/5/7 board from Gigabyte and runs madvr setup without any DPC spikes over 500 us, please PM me.
Thanks

[Wodde]

Try to disable the Intel Turboboost feature in BIOS, also the CPU Powersave Features, like C1.

I think this is the Problem.

[cyberbeing]

usbport.sys
portcls.sys
ataport.sys
scsiport.sys

usbport.sys (USB controller or any driver which interfaces with USB)
Hotfixes of interest: http://support.microsoft.com/kb/981214 OR http://support.microsoft.com/kb/2484031
Try a different USB ports.
Go out and buy a cheap USB add-in card and try that. If it doesn't help, return it. :devil:

portcls.sys (Audio port interface, so any driver which deals with sound or audio)
Hotfixes of interest: NONE
Disable the on-board audio in BIOS if not in use.

ataport.sys (Anything listed under IDE/ATA ATAPI controllers in Device Manager)
Hotfixes of interest: http://support.microsoft.com/kb/2288487 OR http://support.microsoft.com/kb/2292867
Disable the IDE ports in your BIOS if not in use.

scsiport.sys (Anything listed under Storage Controllers or SCSI & RAID controllers in Device Manager)
Hotfixes of interest: NONE, but http://support.microsoft.com/kb/978462 is a long-shot if you use a SATA CD/DVD drive.
Try a different SATA ports/controllers. Disable any SATA ports/controllers in your BIOS which are not in use. If you use an add-in RAID card, try removing it.

Also try Wodde's suggestion as well as rolling back to an older Motherboard BIOS.

After doing all that without any luck, start your search for magical drivers for your hardware. Start with the drivers shipped on CD with your Motherboard, and move forward from there.

Windows 7 SP1 releases on February 22nd to the public. It may not hurt to be patient and wait until then, to save yourself unneeded trouble if it fixes things.
Of course there is a slim chance that SP1 could make things worse, until new drivers are released. :devil:

[Mikey2]

First off - THANK YOU!! I am absolutely in love with your renderer! I am a video-quality freak and my Pioneer Elite Kuro plasma has never looked better!

I have a quick question: I know a lot about all of this and have a software/hardware setup that is extremely complex. However, since I do not do much gaming, I am unclear on how much 3D-hardware settings are utilized.

I currently have a pair of NVidia 8600GT video-cards in running in SLI. (Once I turned on SLI I now have 0 frame-drops using 4-Tap Spline!) Since it appears that you take advantage of some of the 3D-settings on the graphics card (something lesser renderers do not do) are any of the 3D-settings in the NVidia Control panel applicable and can improve quality and/or performance? (For instance, will adding a profile for MPC and setting the Antialiasing Mode to "Enhance any application setting" or "Override any application Setting" do anything? What about the other settings in the control panel?)

Anyway, thanks again for the exceptional renderer; I hope you continue to work on this. In addition, as Moore's law continues to apply (i.e. peoples' computers continue to become faster and better,) that this renderer is discovered by more people.

MikeY

[pankov]

madshi,
is there a chance to get a version of madVR where all the networking stuff is disabled ... or even better - user configurable?
For me, personally, the increased DPC latency is not a problem at the moment because it's bellow the threshold but I'm a little bit worried that it can get higher (either caused by another software/driver or madVR itself) and then it could lead to problems which I'd like to avoid.

[leeperry]

oops, I guess I misunderstood what the DXVA/DXVA2 support in CoreAVC 2.5 would do: http://forums.corecodec.com/threads/1531-When-will-CoreAVC-support-ATI-STREAM-universal-computing?p=11441&viewfull=1#post11441
The CoreAVC filter will need to be connected directly to the renderer for DXVA to work.
So, technicallly this would allow DXVA in mVR w/ AMD graphic cards, but still no post-processing like w/ CUDA.

PS: http://forums.corecodec.com/threads/1531-When-will-CoreAVC-support-ATI-STREAM-universal-computing?p=12248&viewfull=1#post12248
adding specifics for ATI accel and OpenCL are next.

[nevcairiel]

DXVA always needs renderer support. This will NOT allow DXVA support in mVR, unless madshi adds the required support for DXVA2 (which should be possible, imho - DXVA1 is a whole different beast). Supporting DXVA2 is relatively simple, however still needs to be supported.

But then again, if you want maximum quality, go with software decoding already! :d

[nlnl]

But then again, if you want maximum quality, go with software decoding already! :d

And what about amazing vector adaptive deinterlacing without DXVA2?
I guess that VA deinterlacing is the way to go for maximum quality if the source is video (not film).

[leeperry]

But then again, if you want maximum quality, go with software decoding already! :d
..or use CUDA http://img835.imageshack.us/img835/4285/icecreamd.gif

GPU decoding consumes less electricity than 100% CPU solutions apparently(I've got a wattmeter but didn't bother comparing)...and being able to do crazy post-processing on high bitrate BD is another reason.

But all hopes are not lost, wait another year or two and CoreAVC will support OpenCL.

Indeed DXVA2 deinterlacing rocks, but YADIF in double framerate mode isn't too bad either.

[aichan]

i can't install madvr on windows 7 64bit. i've double click install.bat few times, but madvr button on MPC-HC64 still greyed out..
when i was on windows 7 32bit, there was no problem at all.

please help me on this problem.

[namaiki]

i can't install madvr on windows 7 64bit. i've double click install.bat few times, but madvr button on MPC-HC64 still greyed out..
when i was on windows 7 32bit, there was no problem at all.

please help me on this problem.

Use MPC-HC 32-bit. It runs on Windows 64-bit.

[nevcairiel]

..or use CUDA http://img835.imageshack.us/img835/4285/icecreamd.gif

GPU decoding consumes less electricity than 100% CPU solutions apparently(I've got a wattmeter but didn't bother comparing)...and being able to do crazy post-processing on high bitrate BD is another reason.

But all hopes are not lost, wait another year or two and CoreAVC will support OpenCL.

Indeed DXVA2 deinterlacing rocks, but YADIF in double framerate mode isn't too bad either.


The hardware deinterlacing is the only thing thats really worth using, but luckily i don't watch interlaced TV, and rarely ever encounter other interlaced material. :)

And no, CUDA is not the allmighty solution. Its not a CUDA decoder, it just uses CUDA to access the hardware decoding pipeline in the GPU, so its the same decoder as DXVA uses. And those hardware decoders are not always producing 100% perfect quality. Software is the way to go if you really want the last % of quality. Not saying the hardware decoders are necessarily bad. tomshardware actually just had a comparison of hardware decoder implementations, and NVIDIA overall seemed to come out a bit higher then ATI/AMD, but on par with Intel, fwiw. Its usually all right, its just not the 100% :)

But sure, if power or freeing the CPU for other tasks is one of your goals, then it surely is useful

[leeperry]

Its not a CUDA decoder, it just uses CUDA to access the hardware decoding pipeline in the GPU, so its the same decoder as DXVA uses. And those hardware decoders are not always producing 100% perfect quality. Software is the way to go if you really want the last % of quality.

But sure, if power or freeing the CPU for other tasks is one of your goals, then it surely is useful
Well, I've watched my good share of h264 movies w/ CoreAVC CUDA...I even reported some bugs(that were promptly fixed), I can assure you that I've never encountered a single PQ problem w/ CoreAVC CUDA. CoreCodec take forever to release new features, but their work is solid. Nothing to do w/ the DXVA random pixely garbage in MPC-HC.

Electricity is getting more and more expensive everywhere, paying less for our hobbies is something everyone should be interested in...CoreAVC CUDA does just that, just like mVR: they finally put those superfast GPU's to good use...and at what they do best :)

The last two years have been a milestone for HTPC entertainment IME...never been happier :cool:

Now, I really wish mVR would support compressed LUT's together w/ a user friendly way to switch between them...I know it's on madshi's todo list, time will tell whether that'll ever happen.

[ajp_anton]

How does software decoding give better picture quality than hardware decoding?
Apart from errors from streams that aren't supported by the hardware (in which case it should fall back to software anyway), the image should by standard be identical in all decoders.

[aichan]

Use MPC-HC 32-bit. It runs on Windows 64-bit.

wow..it works..
thanks!

[dansrfe]

There is no such thing is getting "better quality" hardware vs. software decoding. Now afaik DXVA decoding decodes only to NV12 or some odd colorspace like that. If it keeps it YV12 and feeds that to madVR then the decoding output should be exactly the same as say ffmpeg_mt software decoding.

[STaRGaZeR]

And what about amazing vector adaptive deinterlacing without DXVA2?

You can use VA hardware deinterlacing AND software decoding without issues ;)

[nlnl]

You can use VA hardware deinterlacing AND software decoding without issues ;)

And so we can use MADVR + VA hardware deinterlacing ?
Please tell me how

[nevcairiel]

How does software decoding give better picture quality than hardware decoding?
Apart from errors from streams that aren't supported by the hardware (in which case it should fall back to software anyway), the image should by standard be identical in all decoders.

"Should" being the important word.
Especially AMDs Motion Compensation has some issues with high motion scenes combined with panning cameras.
There is a fully defined standard for decoding, yes, but that doesn't mean everyone strictly follows it.

Now afaik DXVA decoding decodes only to NV12 or some odd colorspace like that. If it keeps it YV12 and feeds that to madVR then the decoding output should be exactly the same as say ffmpeg_mt software decoding.
NV12 and YV12 are really the same, just differently arranged. You can convert between those two formats without any data loss.

Oh also, again the "Should"
And another thing: ffmpeg-mt has some bugs still that can degrade picture quality. For full software reference, use normal ffmpeg.

And so we can use MADVR + VA hardware deinterlacing ?
Please tell me how

Not possible.

[Luv]

And so we can use MADVR + VA hardware deinterlacing ?
Please tell me how

I might be confused about what's really in motion here but if I watch 1080i material with automatic deinterlacing and Pulldown detection checked in the CCC,I have extraordinary good results with madVR 0.36.

[namaiki]

I might be confused about what's really in motion here but if I watch 1080i material with automatic deinterlacing and Pulldown detection checked in the CCC,I have extraordinary good results with madVR 0.36.

What program are you using? If MPC-HC, could you please tell me what is listed in the menu (Play> Filters) when playing that 1080i material?

[Luv]

namaiki, I use:
- ZP 5.02 on XP/32
- Catalyst 10.8
- latest DirectX (Does everybody pay attention to this ?)
- Powerdvd MPEG2 decoder
- madVR 0.36 (Spline64,4 taps all the way)

[nevcairiel]

madVR tells the graphics card that its a game, not a video player. All those post-processing options built into the graphics drivers should not turn on. Are you sure that there isn't any other filter doing deinterlacing along on the way?

[Luv]

I'm positive,nevcairiel.
I never installed the ZP ffdshow package.I installed first all the filters I wanted and then let ZP detect them when installing itself.

[STaRGaZeR]

And so we can use MADVR + VA hardware deinterlacing ?
Please tell me how

No, with madVR you can't. But you can do it with EVR (or any of MPC-HC's variants) if you feed it with NV12 and you set the proper flags with ffdshow or CoreAVC. CoreAVC puts wrong flags with EVR thou, so if you want this method use ffdshow. I've used it for years without issues.

[Jorgosch]

Hello all,

I've installed 0.36 on top of an older version (I've unregistered and re-registered the filter) but now I have a problem with the quality settings. They just do not stick. Every time I open a video in MPC HC, I have to set the up- and downsampling filters to what I like best because they are stuck at values that don't work well with my weak NVIDIA ION chipset.

I'd like to know where the settings are saved, and if there is a way to manually save my preferences?

Thanks in advance!

Edit: NM, I found in a previous post that I had to run MPC HC as admin.

[ikarad]

Is it possible to add red and green curve with ctrl+J function like in other renderer (evr custom or evr sync)?

[Neeto]

I'm positive,nevcairiel.
I never installed the ZP ffdshow package.I installed first all the filters I wanted and then let ZP detect them when installing itself.

Last time I looked at Power DVD it had options to do de-interlacing in software or hardware.
Is it possible that the Powerdvd MPEG2 decoder is doing the de-interlacing? Just thinking out loud.

[mr.duck]

There is no such thing is getting "better quality" hardware vs. software decoding.
This is a good time to bring up my findings. Using EVR, MPEG2 50i, vector adaptive deinterlacing. With MP-HC software decoder it looks OK, but with Microsoft DTV decoder is is even better for sure. I tried so long to convince myself it is just the same but when I went back to the MS decoder it hit me how good it was. "Ahh that's better!" I would have thought. I cannot get software decoding quality to match DXVA decoding quality no matter what I try. Arcsoft and Cyberlink decoders are as good as MS if you set the options correctly. But MS has no options, it just works. So no need to install others as it comes with Windows 7.


And no, you can't get MadVR and hardware deinterlacing working. In this thread a long time ago I posted screenshots and everything and everyone said it's not going to happen. My opinion is DXVA is needed for the best quality. At least with Radeon graphics cards with their post processing and very high quality deinterlacing. Some 1080i TV is actually progressive, not interlaced. That could be why it looks deinterlaced?

[nevcairiel]

Sounds like you're using some wild image enhancements algorithms like dynamic contrast and color enhancement (hardware decoding usually has those on by default). The software decoders produce a 100% correct decode of the files. Anything else is in the post-processing (which you can of course use when doing software decoding, just not with madVR)

[mr.duck]

Nope I turn all that crap off. Or if you don't install the control panel it is off by default.

When using DXVA the improvement is mainly how fluid the motion is as it's a 50i source being deinterlaced to get 50 FPS. Scrolling text is so clean around the edges and fluid as it scrolls across the screen. But all the other motion in the video looks better too. Like peoples hands, etc. As if it is using a better deinterlacing mode. I'm not sure of the explanation, but there is a difference. Maybe when using DXVA, the video card hardware has better access to the video or something like that? Like I said I don't know why, just that I can't go back now DXVA looks that much better.

[nevcairiel]

The video card doesn't care where the image comes from. Getting software decoding working perfectly with hardware deinterlacing is a bit tricky though, so maybe you just never got it working quite right.
Personally, using yadif with frame doubler and then feeding to madVR is pretty good to my eyes.

[mr.duck]

I tried everything I could. Even with FFDShow and ticking "set interlaced flag in output" or some such option. Nothing was quite as good as the MS DXVA decoding. Software deinterlacing in FFDShow was the worst quality by far that I tried. But yadif with frame doubler was the best of them IIRC. There's not a whole lot of options to change in the MPC-HC software decoder so not much to go wrong. Apart from "Set planar YUV..." option which I'm not sure what it does, I'm sure everything else is set correctly.

Maybe it is only people with Radeon cards trying to get vector adaptive working to full quality. There was a noticeable upgrade in deinterlacing quality when I went from nvidia to AMD.

[dansrfe]

I have no clue why people just don't deinterlace in AviSynth on the fly. It works extremely well, especially well if you are using high quality filters and to around with setmtmode and the caching mechanisms.

[Ger]

I tried everything I could. Even with FFDShow and ticking "set interlaced flag in output" or some such option. Nothing was quite as good as the MS DXVA decoding. Software decoding in FFDShow was the worst quality by far. yadif with frame doubler was the best IIRC though. There's not a whole lot of options to change in the MPC-HC software decoder so not much to go wrong. Apart from "Set planar YUV..." option which I'm not sure what it does, I'm sure everything else is set correctly.

Maybe it is only people with Radeon cards trying to get vector adaptive working to full quality. There was a noticeable upgrade in deinterlacing quality when I went from nvidia to AMD.

My experience is that the MPC-HC MPEG-2 software decoder outputs YUY2 to the EVR renderers in most cases (check the pin info), which explains why you're not getting Vector Adaptive to work, as it's only supported with NV12 as you can see in DXVAChecker. This was pointed out in the MPC-HC thread recently, and the way I understood JanWilhelm, it's on his todo list to increase NV12 merit. I haven't tried his last test builds so I don't know if he's done it yet.

Also, when you tested ffdshow with "set interlaced flag...", did you force NV12 as well?

Another option in the ffdshow output page is to use "Force bob" rather than "Auto". Some interlaced streams have interlaced frames badly flagged as progressive. The "Auto" option relies on correctly flagged frames. Field order can also be wrongly detected in some cases and has a manual override there as well.

[italospain]

I have no clue why people just don't deinterlace in AviSynth on the fly. It works extremely well, especially well if you are using high quality filters and to around with setmtmode and the caching mechanisms.

And i have no clue how to do it right with avisynth can you give me an push to the right direction. What high qualtity filters, caching mechanism ? is it faster than yadif double framerate ?

thanks

[mr.duck]

My experience is that the MPC-HC MPEG-2 software decoder outputs YUY2 to the EVR renderers in most cases (check the pin info), which explains why you're not getting Vector Adaptive to work, as it's only supported with NV12 as you can see in DXVAChecker. ...

Also, when you tested ffdshow with "set interlaced flag...", did you force NV12 as well?

Ahhh that explains it, thanks. I knew that the MS decoder was outputting NV12. I tried to force FFDShow to do that but it seemed impossible. Yes, I'm pretty sure the software MPC-HC MPEG2 decoder does output YUY2. I don't have FFDShow installed any more. I see no need for it. It is hard to set up and doesn't offer any advantages in using it that I can see.


Oh and another reason for wanting DXVA is it locks the Radeon's GPU and memory clocks to a preset middle-ground. If you use software decoding, with full screen video, it spends most of the time at full speed clocks which will increase power consumption.

[dansrfe]

It's just like writing an AviSynth script...but in ffdshow.

Steps to do deinterlacing in ffdshow:

setmtmode(2)
deinterlacer()

AviSynth Video Source checked, Buffers checked (press current a couple of times while the video plays), 3:2 apply pulldown. Make sure you have all the plugins in the plugins directory in AviSynth. For yadif you will need to load it manually with a load_stdcall_plugin line and you should ideally store it outside of the autoloading plugins folder. Use an MT Version of AviSynth like this: http://www.mediafire.com/file/kjdzqyz4dyj/AviSynthMT_258.exe

[pie1394]

Some post-processings NEED be done on the originally decoded 4:2:0 image resolution(NV12, YV12, I420) to get best quality. It includes Deinterlace , Deblocking , Noise reduction and Image Frequency analysis (for Super-resolution up-scaling later).

After that, up / down-scaling, edge smooth / enhancement, Dering just can be applied.

Video Decoder output with 4:2:2(YUY2 / UYVY) means the Chroma U / V components have been vertically up-scaled by 2x.

Vector-Adaptive (or Pixel-Adaptive ... whatever) deinterlace requires huge compuation power and memory bandwidth. It is never heard any General-purposed processor like x86 / PowerPC is capable to do that in real-time, especially for 1080i50 / i60 images. That is why only nVidia / ATI high-end GPU models among all generations can handle that.

[italospain]

NV12 and YV12 are really the same, just differently arranged. You can convert between those two formats without any data loss.


If that is true madVR schould really support NV12 to give the user the possibilty to use Hardware deinterlacing.

If you dont have a high end PC it is the only way to deinterlace HD Material.

[nevcairiel]

You don't get deinterlacing just by accepting NV12. There is more going on, having the image in NV12 is just one of the requirements.

[madshi]

Hi all, I just tested this renderer and in the short time I've seen it, it looks great. I just wanna make sure I've set everything up correctly.

HD 3870 card > ATI DVI/HDMI dongle> HDMI cable into Panasonic G10 plasma
Pixel format set to full RGB in CCC
Using Coreavc for decoding(input and output levels set to 16-235 and BT.709 colorspace selected as I only watch HD)
madVR selected as output in MPC-HC
madVR settings > TV levels selected

Is that OK?
Looks ok to me. The last item on your list "madVR -> TV levels" depends on how your plasma is setup, though. I think you can switch your plasma between TV and PC levels. So you need to make sure that the levels are properly matched. Any calibration disc should allow you to verify that.

While the 24fps --> PAL can lead to a 4% pitch change (pitch, length etc), in the Blu world the 23.976 to 24fps difference is only 0.1%. So while I have no reference to point to, I'd be surprised if the studios would bother with pitch correction OR that we humans would even notice such a small pitch change anyway.
I have been told that pitch correction is not used at all when going from 24.0 to 23.976. The difference is so small that the studios don't feel it necessary to perform a lossy operation like that on the audio.
IMHO you guys have way too much confidence in the studios. I've seen so much "funny" (not really) errors in studio Blu-Ray authoring that I don't have any blind trust in them always doing the "right" thing. I think the chance is pretty high that at least some (though maybe very few) Blu-Rays have been slowed down from 24.000 to 23.976 by using pitch correction. But I have no proof, of course.

I have yet to find any PAL content that has required the use of timestretching though. Perhaps I've just been lucky in my choice of content. The only time it's probably necessary is if the content is natively 25fps rather than a 24 to 25fps conversion.
It depends on the country. E.g. I've been told timestretching is the default method used in DVDs released in Spain. Also I know for a fact that the LOTR DVDs were slowed down by timestretching (at least in Germany).

With a multiple of 23.976 madVR says 1 frame drop every 6.07 days (or a strange garbage number 1.#J days - what's up with that madshi?).
Probably some math overflow. I would see it as an indication of "too good to be true". Meaning your setup is pretty much perfect.

madshi, I've been watching BSG BDs from the discs (the m2ts files direct), using the latest MPC-HC splitter as the Haali Splitter doesn't "see" the DTS-MA stream. The results were basically perfect, no jumps/judders/repeats. Perfectly smooth.

But the last episode is seamless branching, so I had to convert it to MKV, which I then watched using the latest Haali Splitter. Not so good. A couple of jumps - and the feeling of rock solid smoothness was gone.
I've no explanation for that.

The whole 24 / 23.976 stuff is full of myth. Film is 24fps, but both the video and audio are slowed down slightly to 23.976 on Blu to be backwardly compatible with NTSC CE Equipment. Please keep in mind that there is no conversion between 23.976 and 24 itis just a reference to the speed the content is played at.
Careful here. What you say is true for video. That's why eac3to can mod a 25fps video track to 23.976fps without reencoding. *However*, what you say is not true for audio. A 24.000fps audio track plays faster than a 23.976fps audio track and there most definitely *IS* a conversion necessary to convert audio between 23.976 <-> 24.000.

This means that a 23.976 Blu-Ray can not be played back at 24.000 without (audio) conversion. And a 24.000 Blu-Ray can not be played back at 23.976 without conversion, either.

The 23.976 or 24 frame rate is just a number in the header of the file, it has nothing to do with the number of video frames in a file or how it was encoded. Each frame is a single complete frame and the frame rate is just the instruction on what speed to play them back at.... eg in this example the frame rate tells the renderer to display each frame for either 1/23.976 OR 1/24th of a second. For example, I could take these same stream and play it at 50fps but the result would be it would all be in fast motion, sound would be a higher pitch etc.
No. Changing the video speed has zero effect on the audio. If you modded a 24fps track to 50fps, the audio would not change at all. Video and audio would just run very very async.

The problem here is that some people are insistent on bitstreaming audio to have a little TrueHD/DTS HD light turn on with their amplifier, in the belief that it means they are getting better audio quality.

The reality is that bitstreaming is detrimental to audio/video playback from a PC, and that the effects of jitter when comparing LPCM to Bitstreaming with non-PC sources are highly suspect, to say the least.
There should be no difference in quality between bitstreaming and LPCM transport, as long as you can guarantee that (1) nobody does violence to the LPCM data behind your back and that (2) audio and video clocks match perfectly.

To keep audio and video in sync there are a few things you can do.

1. Sync the video to the audio clock. Because your video clocks are never going to be perfect, this means that you will suffer from dropped/repeated frames which really stick out.

2. Sync the video framerate to the video clock, and sync the audio to the audio clock. This will result in perfectly smooth video and perfect audio playback. The audio will not be in sync with the video however resulting in "lip-sync" problems.

3. Sync the video framerate to the video clock, and sync the audio to the video playback without resampling. This results in perfectly smooth video with no dropped/repeated frames, but then the audio will suffer from either dropped/repeated frames.

4. Sync the video framerate to the video clock for perfectly smooth video playback, and resample the audio to match the video. This avoids dropping/repeating frames and keeps audio perfectly synced to the video. (no "lip-sync" problems) ReClock does this with a high degree of precision and can even upsample at the same time to avoid any perceptible loss in quality.
5. Optimize the video refresh rate so well that you can watch a full movie without needing any video/audio drops/repeats.

6. Use madVR 1.0, once it is ready.

:p

Please excuse the late reply - the mainboard of my primary machine got fried and it took a while to get a replacement.
You seem to have hit the nail on the head with the above statement: 0.36 fixed the delay and is working beautifully now :)

You should hear this every day, but thank you very very much for all the countless hours you are spending on coding and supporting madVR - you are one of those "once in an eLifetime"-personalities who give so much to the community and take so little. The least we can do is thank you.

Some posts I read in this thread make me angry. Demanding people who never did anything for the community or just haters - don't take those guys to heart.
Most of us know that you do all of this for free in your spare time and that the skill level you are coding on could only be matched by very few.

Keep up the good work and have a great week!
-trainspotter

P.S.: The only thing left I wonder about is why I was the only person experiencing said delay.
Thanks for the nice words. And I don't really know why you had the delay to start with. I still think there's some kind of network issue on your PC. But it doesn't matter now.

I want to thank madshi for the creation of madVR, it changed the way i watch media now!
Glad you like it!

madVR changing the framerate to a multiple of refresh rate independently of Reclock is beyond the scope of the project.
I disagree. Changing the video refresh rate is naturally the duty of the video renderer. Anything else makes no sense. Having the audio renderer change the video refresh rate behind the back of the video renderer makes as much sense as having the video renderer change the audio DAC sample rate behind the back of the audio renderer.

Hi, by definition shouldn't exclusive fullscreen mode not allow any other windows to appear over the exclusive image. This isn't the case always. Today I had Ember Media Manager scraping in the background to update my nfo files, and watching a movie with madVR / mpc-hc.

The status updates from Ember Media Manager were semi visible over the fullscreen exclusive mpc-hc / madVR window. CTRL-J confirmed this. Can madVR be forced into not allowing anything appear on top of the fullscreen window, and be forced to not allow any other software to ever bring it out of fullscreen mode.
madVR does try to keep on top. I can only guess that the Ember Media Manager also tries to keep on top. There's no way in Windows to demand "I need to be the very top window forever". You can only say "I want to be very top right now". madVR does that when exclusive mode starts, and it does it again when a different window tries to "overpower" madVR. But if the other window/application is trying multiple times to overpower madVR then at some point madVR may give up.

Can anyone tell me why madVR might be grayed-out for me in MPC-HC? I've tried searching around for an answer but everything I find has to do with 64-bit MPC-HC and I'm using 32-bit.
Make sure you don't move the madVR folder after you've run the install.bat.

I just installed Catalyst 11.1a hotfix today and proceed to make sure I set pixel format to full range rgb. After that I was wondering why the color looks weird. Then I realized this is actually the first time I see full range on my projector, or at least this is what I think. Can anyone else try and tell me your result as well? If this is tell then I feel very cheated. I missed out a lot this half past year and going to have to rewatch a lot of stuffs. :(
Weird. I take it with older driver versions switching between full and limited in the driver pixel format showed no difference? And now there is a difference? ATI finally fixing this would be good news, of course.

Or madVR could support DXVA2 one day... :)
Yes, maybe it will.

[madshi]

AFAIR, madVr takes this value from decoder. So you better look into pin-info of decoders/splitters.
No, the display refresh rate is a measurement done by madVR and it should be very very accurate. The information taken from the pin-info of the decoder is the movie frame rate, not the display refresh rate. There's a hint "says source filter" after one line in the OSD. This one line is the information taken from the splitters/decoder's pin-info.

No,nothing except that the rendering times are multiplied by three when I switch from window to Exclusive (This is when very little resizing is required).
If I downscale 1080p to 1280*720,W to Exclusive rendering time is only multiplied by two...

Reinstalled madvr 0.36,didn't make any difference.What is happening ?
Don't worry too much about rendering times in exclusive mode, as long as you don't *see* a problem during movie watching. Actually with the final exclusive mode rendering path, I will not be able to log any rendering times at all.

I was trying to see the differences between VMR9 renderless and madshi and the screenshots i took using both renderers are identical. How is that possible ? and im sure i changed the renderer proper because of the aero effect disable.
After that i changed the default input codec which was libcodec to mpeg4s dmo and the screenshots i took have a few differences with the vmr9 looking better for me.

I have uploaded the screenshots for u to see here (http://screenshotcomparison.com/comparison/25429/picture:0)

source file used for is a DivX5 avi 640x352
comparison 1 is input libcodec output madshi and VMR9 renderless
comparison 2 is input mpeg4s dmo output madshi and VMR renderless
dont forget to click on #2 and see the second image set

btw.. ofc you cannot watch movie with mpeg4s dmo even if the "still" picture looks better in the screenshot its a total mess on moving, like im watching youtube old videos :p

oh i forgot the question: 1) how can i bee sure that im using madshi (even detail info TAB shows only YV12) and 2) can you suggest me a setup to use with KMP3 that will have the best image quality output to my 1080p monitor? im using a laptop VPC F11M1E
Some people find the difference between madVR and other renderers significant. Others don't see a difference at all. Whether there's a difference visible depends on a lot of things. E.g. how good a job does the other renderer do? Which scaling algorithms are used (if any)? How difficult to render is the video frame? E.g. madVR is known to avoid adding any banding, while most other renderers tend to add (or intensify existing) banding in certain kind of scenes. But this is only visible in very specific types of scenes. Another typical case where differences can be seen is when chroma upsampling problems are visible (e.g. red fonts on black background). Or when large amounts of scaling are used.

In MPC-HC you can right click on the video image and then select "Filters" form the context menu. If madVR is used you should see "madVR Renderer" in the "Filters" submenu.

My i5 2300 sandbridge with onboard vga is sadly not fast enough to support any upsampling, for instance 720p to 1080p i get a lot of dropped frames. I already increased rhe memory to 480 mb but sadly to no avail.

After some experimenting i came to the conclusion that if i switch either chroma upsampling or luma upsampling to bilinear i get smooth playback so clearly the shaders aren't fast enough.

The normal trade quality for performance options do not help at all.

So which will have the largest impact on the picture quality a lower chroma upsampling quality or a lower luma upsampling quality ?
First of all I'd suggest to try fullscreen exclusive mode and to change all "flush & wait" options to "flush". Maybe that helps?

If your GPU is not fast enough for doing both chroma and luma upscaling, I'd suggest to switch chroma upscampling to "bilinear". The quality loss will be low, not visible in most scenes. Luma resampling is much more important.

madshi, can I request that frame repeats be displayed in the Ctrl-J stats.
Can't, with the current presentation logic, which is a general purpose presentation logic, working for all combinations of movie frame rate and display refresh rate. I plan to add a special presentation logic in a future madVR version for 1:1 (or 2:1 etc) maps between movie frame rate and display refresh rate. With that special presentation logic I should be able to log repeated frames, too.

Massive overkill really helps. A real quad core, and beefy gaming card are the go. You need headroom.

I just moved my PC to the next room, and ran the cabling through the wall. Noise is not a problem.

If you want silent, stick with Powerdvd and DXVA.

I know madshi has a silent PC, and I think it limits even him.
I have a silent HTPC, but it has a quad core CPU. What limits me currently is the onboard NVidia 9400 GPU. I plan to add a PCIe GPU sooner or later to overcome this limitation. I'd love a 28nm GPU, though (for best power per watt performance), which will only be released by the end of 2011.

Spline resizing is not lower quality. In fact it is pretty much equivalent to what madVR does from the higher-end resizers.
FWIW, madVR resizes in 32bit floating point per component, while AviSynth scripts work in (or at least output) 8bit integer per component. You might be able to see a difference if you upscale the madTestPatternSource "smallRamp" test pattern to 1080p. With madVR scaling this test pattern should result in a perfectly smooth grayscale. With 8bit integer scaling I think there should be faint vertical lines visible in the grayscale.

Furthermore I'd guess (but I'm not 100% sure) that your AivSynth script will likely convert the image to 4:4:4 first, then upscale, then convert back to 4:2:0 (which is the only format madVR accepts). Doing 4:2:0 -> 4:4:4 -> scale -> 4:2:0 is not really optimal for image quality.

Of course the question is if these differences will be visible during normal movie viewing. I won't be the judge of that.

does madvr do any kind of vsync optimizations to fix tearing?

also, what does exclusive mode do?
madVR tries to optimize vsync switching so that there's no tearing in windowed mode. It works for most people, but not all.

Exclusive mode has a number of advantages, one of which is the guarantee to never have tearing. Exclusive mode has one disadvantage, though: You can't have normal GUI (windows, context menus) etc during exclusive mode. So if the media player tries to show some GUI, madVR will have to switch between exclusive and windowed mode, and the switch is not perfectly seamless.

First off - THANK YOU!! I am absolutely in love with your renderer! I am a video-quality freak and my Pioneer Elite Kuro plasma has never looked better!
:)

I have a quick question: I know a lot about all of this and have a software/hardware setup that is extremely complex. However, since I do not do much gaming, I am unclear on how much 3D-hardware settings are utilized.

I currently have a pair of NVidia 8600GT video-cards in running in SLI. (Once I turned on SLI I now have 0 frame-drops using 4-Tap Spline!) Since it appears that you take advantage of some of the 3D-settings on the graphics card (something lesser renderers do not do) are any of the 3D-settings in the NVidia Control panel applicable and can improve quality and/or performance? (For instance, will adding a profile for MPC and setting the Antialiasing Mode to "Enhance any application setting" or "Override any application Setting" do anything? What about the other settings in the control panel?)
Keep all these settings to "application specified" (or however that option is named exactly). Forcing antialiasing or anisotropic filtering will not help image quality at all. If forcing these options changes anything (not sure on that), it will be for the worse.

is there a chance to get a version of madVR where all the networking stuff is disabled ... or even better - user configurable?
For me, personally, the increased DPC latency is not a problem at the moment because it's bellow the threshold but I'm a little bit worried that it can get higher (either caused by another software/driver or madVR itself) and then it could lead to problems which I'd like to avoid.
What makes you think the latency is caused by the network stuff? The madVR network is initialized during start time and then it sits idle, doing nothing. IMHO the chance that the madVR network stuff is causing increased DPC latency is pretty much zero.

DXVA always needs renderer support. This will NOT allow DXVA support in mVR, unless madshi adds the required support for DXVA2 (which should be possible, imho - DXVA1 is a whole different beast). Supporting DXVA2 is relatively simple, however still needs to be supported.
Agreed.

And so we can use MADVR + VA hardware deinterlacing ?
Please tell me how
Currently not possible - unless some decoder/deinterlacer filter uses the DXVA deinterlacing first and then copies the deinterlaced frames back from GPU RAM to system RAM. I don't think anybody is doing that.

"Should" being the important word.
Especially AMDs Motion Compensation has some issues with high motion scenes combined with panning cameras.
Is there any place where I can read more about this?

I might be confused about what's really in motion here but if I watch 1080i material with automatic deinterlacing and Pulldown detection checked in the CCC,I have extraordinary good results with madVR 0.36.
I don't know who's doing the deinterlacing in that case, but I highly doubt it's the GPU hardware. Unless the decoder is copying the deinterlaced frames back into system RAM, which I think is unlikely.

Is it possible to add red and green curve with ctrl+J function like in other renderer (evr custom or evr sync)?
Yes, it is possible. But I don't really see much sense in that. What would you need that for?

And another thing: ffmpeg-mt has some bugs still that can degrade picture quality. For full software reference, use normal ffmpeg.
Hmmmm... The ffmpeg-mt "todo.txt" doesn't mention any possible picture quality degrading which would need fixing. Are you sure on that? What bugs are you talking about?

And no, you can't get MadVR and hardware deinterlacing working.
Not currently, anyway.

Vector-Adaptive (or Pixel-Adaptive ... whatever) deinterlace requires huge compuation power and memory bandwidth. It is never heard any General-purposed processor like x86 / PowerPC is capable to do that in real-time, especially for 1080i50 / i60 images. That is why only nVidia / ATI high-end GPU models among all generations can handle that.
Does ATI/Nvidia "vector-adaptive" deinterlacing do the same thing that HQV, Gennum and Faroudja are doing? Namely per pixel motion adaptive deinterlacing with vertical filtering? Or do ATI/NVidia do full motion compensated deinterlacing? The latter should be a step up from HQV, Gennum and Faroudja.

[madshi]

Options -> Internal Filters -> MPEG-2 Video -> Tick 'Set interlaced flag in output media type'

That fixed it.
This is just another bug in the MPC-HC decoder. madVR asked the decoder to deliver 720x480 pixels, because that's the size of the physical video. But that made the decoder crash. VMR/EVR always round up to a 2^x width. The next madVR version will do that, too. So madVR will ask for 1024x480. Not really a nice solution, IMHO, because it uses more memory than necessary, but it works around this MPC-HC decoder bug. And maybe a few other buggy decoders, too.