madVR - high quality video renderer (GPU assisted)

[ryrynz]

Originally Posted by therock003

Will there ever be an x64 build for madVR so i can use it with mpc x64?

Quite likely yes, but not anytime soon (probably a couple of years) you'll just have to use MPC x86 instead.

Originally Posted by therock003

Also i have a notebook with IntelHD graphics, and upscaling sd with jinc 3-passes gives seriously low performance. Do i need strong hardware in order to be able to get decent upscale performance?

Yes you do, especially with Anti-Ringing enabled. Use Lanczos 3 taps instead and look to change your Chroma to Bicubic 75 both settings without Anti-Ringing enabled if need be.

[therock003]

Originally Posted by ryrynz

Yes you do, especially with Anti-Ringing enabled. Use Lanczos 3 taps instead and look to change your Chroma to Bicubic 75 both settings without Anti-Ringing enabled if need be.

Well Playback on these settings was smooth and i even had antiringing on with lanczos. But i didnt notice any difference between this and EVR. I had one process of x86 with madVR and the x64 with eVR at the same time, and freezed the same frame and alternate the windows on fullscreen, but i couldnt notice any difference

[e-t172]

Originally Posted by 6233638

I think you have it the other way around - without knowing the black level, you cannot use a black level compensated gamma value.

No, actually, I stand by what I said. The calculator results are expressed in terms of absolute luminance, as well as BT.1886. However, madVR is unable to target some absolute luminance because it doesn't know what the luminance of 0% and 100% is. When you specify a pure power curve in madVR, it is applied in terms of pixel values, not luminance. So in the end, the output of madVR, after the pure power curve has been applied, is offset by the monitor's black level (madVR pixel value 0 = monitor black level). And when you offset a pure power curve with the monitor's black level, you get… a black-compensated power curve. QED.

The important thing here is that when configured in madVR a pure power law is applied to pixel values, but luminance ≠ pixel value. Instead, luminance = a*(pixel value)+b. The resulting luminance follows a black-compensated curve, even though the pixel values follow a pure power curve.

Originally Posted by 6233638

This post is a pretty good indicator of why BT.1886 is better than a black level compensated power curve though. Now that BT.1886 exists, I don't see a reason to calibrate to anything else (for HT purposes at least) if you have the ability to.

I totally agree, but considering very few calibration solutions support BT.1886 right now, most people won't have this ability. A good first step would be to have BT.1886 support in madVR, so that BT.1886 can be used even with a monitor using a different curve (e.g. black-compensated pure power law, sRGB function).

[ryrynz]

Originally Posted by therock003

i didnt notice any difference between this and EVR.

If you take a screenshot of each one and enlarge a portion of it you'll be able to more easily see the difference, it may be hard to notice the differences on a "small" screen especially to an untrained eye.

[therock003]

Originally Posted by ryrynz

If you take a screenshot of each one and enlarge a portion of it you'll be able to more easily see the difference, it may be hard to notice the differences on a "small" screen especially to an untrained eye.

Well the point is that the difference in quality should be significant so that you can notice it live as youre watching the video, and note examine a frame in detail to see the better results. Anyway how can a capture a full screen so i can post it, and analyze the differences. I'm using mpc but i think saving an image, it saves it at native resolution and not the upscaled fullscreen image.

[Niyawa]

Originally Posted by therock003

Well the point is that the difference in quality should be significant so that you can notice it live as youre watching the video, and note examine a frame in detail to see the better results. Anyway how can a capture a full screen so i can post it, and analyze the differences. I'm using mpc but i think saving an image, it saves it at native resolution and not the upscaled fullscreen image.

This is true. But as some said, you need a trained eye for that. Some people will see the difference (they claim it's around 6-10%) others will just say it's not worth it. In any case, I've been using madVR for more than a year and a half now. Whenever I see EVR-CP I can somewhat see a blurry or distorted red, it seems like the colors are not vibrant as they should. Maybe this is my own self-inserted placebo effect, but it happens. The real thing really changes when we make upscaled screenshots comparison though.

You can use Print Screen and paste the copied image in a image editor, Paint will do. Save in .png of course.

[madshi]

Originally Posted by italospain

Thanks for the explanation.
But wouldn't it possible to "talk" with LAV Filters ?

For what exact purpose?

Originally Posted by fallengt

It's always unchecked, doesn't change anything. Also whenever I switch to fullscreen exclusive mode, my GPU(ATI HD6850) load skyrocket to 80%.
I think I messed up somewhere, not sure what it is though.

That sounds weird. Maybe you could try uninstalling/reinstalling the GPU driver, or maybe there's a way to reset the GPU driver to default settings?

Originally Posted by leeperry

allow me to ask whether the AR filename tag is on your current todo list ?

Yes.

Originally Posted by turbojet

With extended displays EVR has the same problem and it uses the primary displays composition rate. Disabling composition fixes this, is there a way to ignore composition in this type of situation with madvr?

If composition is enabled (and it can't be disabled in win8) and if you're not using overlay or FSE, then there's no way to "ignore" composition. Direct3D is redirected to composition internally in the OS in that situation. No way around it, except by disabling composition, using overlay or FSE.

Originally Posted by turbojet

However switching from one display to another with different refresh rates, evr handles fine, madvr doesn't and it uses the first displays composition rate.

If EVR can handle this then it might be considered a madVR bug. So please create an entry in the bug tracker, with an attached log file and a detailed description how to reproduce it (how to setup dual monitor so that the issue occurs etc). I can't promise I'll look at this soon, though, because multi-monitor issues are really hard for me to debug cause my development PC doesn't have a dual monitor setup. Please also write into the bug tracker description whether the problem does not occur if you directly start a new media player instance on the target monitor. (Probably it will work fine then?)

Originally Posted by turbojet

Here's a log for display changer not working after using cru. nvidia CP and nircmd works fine.

According to the log your monitor has no display modes listed to switch to. My best guess would be that the cru utility results in madVR detecting a new display. Check the device stuff in the madVR settings.

Originally Posted by turbojet

edit2: OTOH if aero peek worked with overlay I'd use it most of the time. Is this possible?

You'd have to ask Microsoft about that.

Originally Posted by RamGuy

Does anyone have any idea how a Intel i5-4570R, i5-4670R or a i7-4770R with the new Haswell HD 5200 (GT3e) will perform with MadVR? Might consider a Mini-ITX Media Center running Media Browser + MPC-HC if the performance is good enough.

madVR already works pretty well with Intel HD4000 GPU. Of course there are limits, e.g. it definitely won't do 60p Jinc3 resizing etc. But if you're willing to make a few compromises with the scaling algorithms, it should work very well. The Haswell graphics will likely be a bit faster than HD4000, but not worlds faster, so I believe you'd still have to accept some compromises with Haswell, as well. Haha! With "[H][as][well] as well". If you don't want to make any compromises you'll likely have to go with a NVidia 660 or something with similar power from AMD. And even that is only with the current madVR algorithms. I don't know whether maybe in the future there'll be new algorithms which consume even more power.

Originally Posted by iSunrise

If I have the option to choose an extremely close approx. refresh with very small deviations on my monitor to the native frame rates like 23.976fps, 25fps or 29.970fps AND also a multiple for all of them, what would I go for?

What looks best to your eyes. There's no general answer to your question, it all depends on the exact hardware (GPU and display) you're using. Simply try both and trust your eyes.

Originally Posted by chuuey

Well I think my Vsync issues on XP are a problem related to the Nvidia drivers. Even with latest WHQL drivers, the refresh rate reported by madVR is all over the place, sometimes falling to 57Hz on a 60Hz screen and sometimes jumping into 62 range, and frame drops occur everywhere. I just tested madVR on another machine with Intel HD4000 Ivy Bridge graphics, exact same software setup as my PC and it works flawless, no Vsync issues and I just watched a film with 0 frame drops on it. I guess Nvidia messed something up on XP and 600 series drivers. Interesting fact is that all games, for example have perfect Vsync applied.

Ah, XP. I don't really know what's going on, nobody else except you seems to have this specific problem. It could be a driver problem, but then why does nobody else suffer from it? I believe there are still several people on XP (although the number is decreasing). It's quite possible that your hardware has a problem. You could try getting a comparable GPU from a different manufacturer, or even the same GPU from the same manufacturer, and then sell your current GPU on ebay. It could also be a problem related to XP. Maybe reinstalling XP would solve the problem, but doing that is no fun, of course.

Anyway, there's really not much I can do. It's pretty clear that the vsync scanline information your GPU/driver reports are very unreliable. Games are not using that information, but madVR needs it.

Originally Posted by jmelan

I have a GTX 650 Ti and noticed that the included madNvLevelsTweaker.exe utility has the added benefit of fixing the Nvidia HDMI audio "not plugged in" problem after resume from sleep/TV turned off and back on (normally requires a restart).

Really?? I'm surprised...

Originally Posted by jmelan

Not knowing how the madNvLevelsTweaker.exe utility is working, is it safe to run multiple times (eg every time I resume from sleep)?

Yes, it should be safe.

Originally Posted by dbcooper

Hi guys, have a quick question about DVD playback.

Many PAL DVD's are 25fps progressive but flagged as 50i, where the frames are progressive (at 25fps) but are played twice to give 50fps.

Can I simply deselect "automatically activate deinterlacing when needed" and safely have the video played as 25p?

For most PAL DVDs deactivating deinterlacing should result in perfect 25p. However, there are a few odd DVDs where parts of the movie, or even the full movie would be full of weaving (interlacing) artifacts this way. The best solution would be to force madVR into film mode deinterlacing. That should always result in perfect 25p output - as long as the movie is true film content. There is some content out there (especially anime) which is field blended, though. For such content you can't use madVR's film mode, nor can you deactivate deinterlacing, for such content you need to use full video mode DXVA deinterlacing.

Originally Posted by adhara

What a pity.
I noted (MadVR + Lav (Jriver)) the following distribution in terms of use of machine resources (in average and with Bicubic/Lanczos).

GPU (hd4000) : 60%-80%
CPU (i7 3370): 5%-10%

Many people are moving towards a fanless configuration (w/o any graphic card).

You can't have a PC without any GPU. Such a PC would not be able to drive any monitor. What you probably mean is that many people are moving towards a PC with a built in GPU (e.g. Intel's HD4000 GPU).

Originally Posted by adhara

It would be nice that developers balance their codes load between CPU and GPU.
Currently, most of codes are GPU oriented. Not sure this is the better choice.

CPU and GPU are very very different designs. The CPU is ideal for some algorithms, but not so good for other algorithms. Same with the GPU. All algorithms where every pixel runs through the same calculations are perfect for the GPU and would run rather slowly on the CPU. E.g. Jinc3 scaling would never run in full speed on a GPU, I believe, not even on the fastest (consumer) CPU available.

My personal opinion is that video decoding should ideally be done via CPU and then all video image algorithms should run on the GPU. And that's a solution which is absolutely possible with madVR to achieve. The current integrated GPUs are still relatively slow. But their performance is increasing much faster compared to the increase in CPU performance. So a few years from now GPU performance will probably be multiple times that of today. So in the long run the design chosen for madVR will play out to be the right one.

Maybe someday I'll offer to run some parts of madVR on the CPU instead of the GPU, but in order to do that I'd have to rewrite every algorithm for the CPU. So it would be a LOT of work. So if it ever comes it will be a long time from now.

Originally Posted by totalz

I'm using LAV video decoder + MadVR, and am wondering which hardware decoding setting to choose in LAV.

With MadVR, which hw-decoding is recommended? DxVA2 native or copy-back?

Personally I recommend software decoding, if your GPU is fast enough. If you want to use hw decoding, "native" should probably perform faster than "copy-back".

Originally Posted by leeperry

BTW, could the MPEG1 chroma be properly aligned with a simple PS script?

That should be possible, although it might be ever so slightly lower quality then if done right in the first place.

Originally Posted by therock003

Well the point is that the difference in quality should be significant so that you can notice it live as youre watching the video, and note examine a frame in detail to see the better results.

madVR does have a number of advantages over EVR, but not all of them are obvious. Some image quality advantages only show in certain kinds of scenes. E.g. differences in chroma upsampling show well with scenes where there's lots of contrasty red and black stuff. E.g. red fonts on black background. Or advantages in color processing bitdepth and dithering only show if you have content with very smoothly blended gradiants. Or advantages in scaling algorithm only show if you actually use scaling, and hopefully more than just by a small amount. The higher the scaling factor, the bigger the differences. Or the madVR anti-ringing filter can only show its benefits if you scale videos which have high contrast edges.

Maybe some day I'll provide a collection of test samples which demonstrate where madVR beats EVR noticeably. It's a bit difficult, though, because if I use Hollywood movie samples for that, I could run into legal trouble.

madVR also has some other advantages over EVR. E.g. if your display can only do 60Hz, but doesn't support 24Hz, then only madVR allows smooth playback (without 3:2 pulldown judder) of Blu-Ray movies on your display. EVR can't do that. Or madVR has a better seeking behaviour. Or madVR has a built in refresh rate switcher. Or madVR has the best IVTC algorithm built in that is available in the HTPC world (as far as I know).

Originally Posted by hdboy

For some reason, I am getting huge number of frame drops playing some trailers with smooth motion on (24fps->60).

Don't see frame drops here with my Intel HD4000 GPU. Which queues are near empty when those frame drops occur? Have you tried in fullscreen exclusive mode?

[DragonQ]

Originally Posted by madshi

Or madVR has the best IVTC algorithm built in that is available in the HTPC world (as far as I know).

Well there's surely room for improvement, considering almost all of my 25p-inside-25i streams cause MadVR to activate deinterlacing - it's fortunate that my GPU handles this properly. I don't think I've ever seen MadVR activate IVTC automatically, unless I'm just misunderstanding the OSD.

[nevcairiel]

Originally Posted by DragonQ

Better than those found in GPUs though?

The ones in GPUs are rather limted, you can be lucky if it handles 3:2 properly, any other cadence will most like just not work - and not to mention that it doesn't decimate after the IVTC, so you can't get 24p out of a telecined 29.97 fps movie.
So yes, much better.

The only thing missing is auto-detection, so i don't need to manually force it.

[DragonQ]

Quicker than a flash...

[HitomiKun]

Every video file I open right now causes madVR to crash. I have sent a few reports through the bug reporting screen that comes up every time madVR crashes. I am using a mpc-hc fork from JanWillem32, the latest mpc-hc x86 build with SSE2 instruction set. See his development folder for the builds:
http://www.mediafire.com/?xwsoo403c53hv

Other than that, I use the latest versions of madFlac + LAV Filters + AC3 Filter to play my videos. I used to have Haali on the list, but LAV Filters got Ordered Chapters support for MKV files, so I dropped Haali and uninstalled it.

Setup looks like this:

all set to "prefer"

This is why I'm using JanWillem32's builds:


I recently switched to Windows 8 Pro 64 bit and just installed my good old madVR setup for mpc-hc and it didn't work as I started the first file. (This means I haven't got the chance to do settings for madVR, since I always started a video file to access madVR settings)

[madshi]

Originally Posted by DragonQ

Well there's surely room for improvement, considering almost all of my 25p-inside-25i streams cause MadVR to activate deinterlacing - it's fortunate that my GPU handles this properly. I don't think I've ever seen MadVR activate IVTC automatically, unless I'm just misunderstanding the OSD.

What you're talking about here is:

(1) automatic detection when deinterlacing is needed
(2) film vs video mode detection

For (1) madVR simply uses the video stream flags at the moment, which is somewhat unreliable. For (2) there's no code in madVR at all yet. I plan to write new algorithms for (1) and (2) in the future to do everything automatically with hopefully near perfect reliability. Of course the new algorithms will eat additional CPU and/or GPU resources.

Originally Posted by nevcairiel

The ones in GPUs are rather limted, you can be lucky if it handles 3:2 properly, any other cadence will most like just not work - and not to mention that it doesn't decimate after the IVTC, so you can't get 24p out of a telecined 29.97 fps movie.

FWIW, at least my AMD GPU does detect some other cadences, too, but it's all not very reliable. E.g. it often uses video deinterlacing for PAL movie content etc...

Originally Posted by nevcairiel

The only thing missing is auto-detection, so i don't need to manually force it.

Exactly. Oh well, probably support for mixed video + film content will also need to be added which could be quite difficult. <sigh>

[madshi]

Originally Posted by HitomiKun

Every video file I open right now causes madVR to crash. I have sent a few reports through the bug reporting screen that comes up every time madVR crashes. I am using a mpc-hc fork from JanWillem32, the latest mpc-hc x86 build with SSE2 instruction set.

Unfortunately you don't have the MPC-HC debug symbols (*.pdb) installed. Which means that the bug reports don't include information of what MPC-HC was doing exactly when the crash occurred. From what I can see the crash occurs inside of madVR always at the same place. But the place makes no sense at all. My first guess would be that MPC-HC damages something which results in madVR crashing. But that's nothing but a wild guess. Does the same problem also occur with normal MPC-HC or MPC-BE builds?

[HitomiKun]

using a normal build from
http://xhmikosr.1f0.de/mpc-hc/
does not crash.

[mark0077]

Originally Posted by madshi

What you're talking about here is:

(1) automatic detection when deinterlacing is needed
(2) film vs video mode detection

For (1) madVR simply uses the video stream flags at the moment, which is somewhat unreliable. For (2) there's no code in madVR at all yet. I plan to write new algorithms for (1) and (2) in the future to do everything automatically with hopefully near perfect reliability. Of course the new algorithms will eat additional CPU and/or GPU resources.

madshi, quick question. Do you ever see it being possible or useful for parts of madVR like this detection algorithm to become their own product or filter, outside of a video renderer. I'm sure many like myself are very interested in some of these amazing improvements and algorithms but because I use filters like ffdshow / SVP between the decoder and renderer, I have to basically disable or not use any of these features, because the renderer is too late in the chain to do this detection / deinterlacing. If they were more modular I imagine being able to use them in whatever part of the chain would suit, like after decoding but before SVP for example? I'm just afraid that if they are all contained in one big place, ie madVR, that many of its features can't be used by others that use filters before the renderer.

[madshi]

Originally Posted by HitomiKun

using a normal build from
http://xhmikosr.1f0.de/mpc-hc/
does not crash.

Then I would guess that the crashing MPC-HC build is probably at fault.

Originally Posted by mark0077

madshi, quick question. Do you ever see it being possible or useful for parts of madVR like this detection algorithm to become their own product or filter, outside of a video renderer. I'm sure many like myself are very interested in some of these amazing improvements and algorithms but because I use filters like ffdshow / SVP between the decoder and renderer, I have to basically disable or not use any of these features, because the renderer is too late in the chain to do this detection / deinterlacing. If they were more modular I imagine being able to use them in whatever part of the chain would suit, like after decoding but before SVP for example? I'm just afraid that if they are all contained in one big place, ie madVR, that many of its features can't be used by others that use filters before the renderer.

Sorry, but in the long run I plan to run all those algorithms on the GPU, anyway (currently IVTC is still done on CPU, though), which means that there's no way other filters can be executed in between.

[daltonm]

Originally Posted by hdboy

For some reason, I am getting huge number of frame drops playing some trailers with smooth motion on (24fps->60). For example, this one

http://videos.hd-trailers.net/20130409_elysium_trailer1_4000.mp4

mediainfo tab

General
Complete name                  : S:\20130409_elysium_trailer1_4000(2).mp4
Format                         : MPEG-4
Format profile                 : Base Media / Version 2
Codec ID                       : mp42
File size                      : 63.0 MiB
Duration                       : 2mn 12s
Overall bit rate mode          : Variable
Overall bit rate               : 4 000 Kbps
Encoded date                   : UTC 2013-04-09 15:15:49
Tagged date                    : UTC 2013-04-09 15:15:49

Video
ID                             : 1
Format                         : AVC
Format/Info                    : Advanced Video Codec
Format profile                 : Main@L5.1
Format settings, CABAC         : Yes
Format settings, ReFrames      : 4 frames
Codec ID                       : avc1
Codec ID/Info                  : Advanced Video Coding
Duration                       : 2mn 12s
Bit rate mode                  : Constant
Bit rate                       : 3 936 Kbps
Width                          : 1 920 pixels
Height                         : 1 080 pixels
Display aspect ratio           : 16:9
Frame rate mode                : Constant
Frame rate                     : 23.976 fps
Color space                    : YUV
Chroma subsampling             : 4:2:0
Bit depth                      : 8 bits
Scan type                      : Progressive
Bits/(Pixel*Frame)             : 0.079
Stream size                    : 62.0 MiB (98%)
Language                       : English
Encoded date                   : UTC 2013-04-09 15:15:49
Tagged date                    : UTC 2013-04-09 15:15:49

Audio
ID                             : 2
Format                         : AAC
Format/Info                    : Advanced Audio Codec
Format profile                 : LC
Codec ID                       : 40
Duration                       : 2mn 12s
Bit rate mode                  : Variable
Bit rate                       : 61.6 Kbps
Maximum bit rate               : 192 Kbps
Channel(s)                     : 2 channels
Channel positions              : Front: L R
Sampling rate                  : 44.1 KHz
Compression mode               : Lossy
Stream size                    : 994 KiB (2%)
Language                       : English
Encoded date                   : UTC 2013-04-09 15:15:49
Tagged date                    : UTC 2013-04-09 15:15:49
mdhd_Duration                  : 132168

Odd because my system can play much more demanding material, like bluray m2ts files, just fine with smooth motion on. I am using mpc-hc and lav splitter and lav decoder. Any idea? thanks

I have the same problem.
look
http://i.imgur.com/zixIBqG.jpg

edit
the problem does not happen with exclusive fullscreen mode.
edit2
fullscreen windowed mode with 4 backbuffer queue
Problem Solved.. LoL
http://i.imgur.com/We2Ptmi.jpg

[HitomiKun]

I'll try to compile JanWillem32's source code, though that'd take a few hours since I have to install visual studio and all that stuff. If you madshi have the right tools to compile his code I'd recommend you try it, would take less time and you'd experience it all by yourself:

his git: https://github.com/JanWillem32/mpc-hc

edit:
just read your post, if it isn't madVR then I'm sorry for the trouble I'll try to use an older version from JanWillem32.

[dansrfe]

I use the ICL3 and either madVR or LAV crashes when I press Alt + F4 in exclusive or windowed mode sometimes. I'm not sure I can replicate the scenario at will but I may be able to get a crash report if I keep using the debug varsion of madVR for some time.

[hdboy]

Originally Posted by madshi

Don't see frame drops here with my Intel HD4000 GPU. Which queues are near empty when those frame drops occur? Have you tried in fullscreen exclusive mode?

Ah, FSE fixes it. But for usability purpose, I like to avoid FSE if possible.

Sorry, how do I tell which queues are near empty?

Changing back buffer queue from 8 to 4 did not help. Reducing CPU and GPU queues all way down to 4 seems to reduce # of dropped frames, but not eliminate them. Increasing CPU and GPU queues to max causes more drooped frames.