madVR - high quality video renderer (GPU assisted)

[Shiandow]

I'm having a problem getting XySubFilter to load with MadVR when ffdshow raw video filter is enabled. Works fine without ffdshow in the stream.


As madshi said this seems to be XySubFilter related, but I happen to recall having a similar issue with ffdshow which could be solved by going to the ffdshow options, then "Subtitles" and then disable everything under "Subtitle sources", in particular "Embedded subtitles". If that doesn't work you should try asking in the XySubFilter thread.

[flashmozzg]

http://blogs.nvidia.com/blog/2014/03/20/directx-12/

Doesn't matter right now. It(dx12) will ship in late 2015, most probably with win9 (it's usually 3 years between major win releases).

[clsid]

With regard to the external DVD subtitle problem, the behavior that I have observed is:
A subtitle that has to be displayed at time X is not displayed when video is at position X but when playback has been active for X amount of time. Seeking has no effect at all.

[kasper93]

1- open a dvd (disc or folder) with mpc-hc;
2- drag a .srt file to mpc-hc window;
result: Subtitles are blinking fast and the timing is wrong;

this problem doesn't happen with vmr9 renderless, for example.

Can anyone confirm this?

Seven x64 with gtx 650 ti @332.21 and mpc-hc (20/03/2014)

yes, also on my computer subtitles are blinking when playing dvd with external subtitles.
I guess madvr doesn't like dvd navigator filter which replaces lav splitter when playing from dvd.

Windows 8.1, intel hd4600, latest mpc-hc & madvr.

which subtitle renderer are you using? Is smooth motion frc on/off? If it's on, does the problem go away if you turn it off?

with regard to the external dvd subtitle problem, the behavior that i have observed is:
A subtitle that has to be displayed at time x is not displayed when video is at position x but when playback has been active for x amount of time. Seeking has no effect at all.

This in know problem with ISR and madVR and how they handle subtitles for DVD or timing in fact. And I totally forgot about it being that I don't watch DVD that much anymore and never with external subtitles :)

This build should resolve this issue https://www.dropbox.com/s/kifqfx7li1w75az/MPC-HC.1.7.3.119.x86_ISR_DVD_TIMING_FIX.7z I noticed few other (very) minor issues, but I will deal with them later.

Also could we move discussion about MPC-HC to its own thread?

[TheShadowRunner]

Hi madshi,
A rather annoying cosmetic change has occured between builds 0.87.4 and 0.87.5.
I'm using the vanilla MPC-HC standalone VSfilter v1.7.3 here
In 0.87.4 and all previous versions of madVR (and all the other video renderers), switching the subtitles language while a video is playing "froze" the image for an extremely small amount of time and subs started appearing.
In 0.87.5 and newer, switching the subtitle language while a video is playing has madVR display a black screen (!), then the image comes back with the newly selected subs.
From the changelog between 0.87.4 and 0.87.5, I believe the change responsible for this behavior could be the following?
*exclusive > windowed switch now shows a black frame instead of an old one
Is it possible to revert this change or don't have it active when VSfilter is used?

[Delerue]

This in know problem with ISR and madVR and how they handle subtitles for DVD or timing in fact. And I totally forgot about it being that I don't watch DVD that much anymore and never with external subtitles :)

This build should resolve this issue https://www.dropbox.com/s/kifqfx7li1w75az/MPC-HC.1.7.3.119.x86_ISR_DVD_TIMING_FIX.7z I noticed few other (very) minor issues, but I will deal with them later.

Great! It seems perfect to me now. Thanks.

Also could we move discussion about MPC-HC to its own thread?

For sure. I thought it was a madVR problem since it doesn't happen with other renders but madVR. Sorry. :)

[MistahBonzai]

As madshi said this seems to be XySubFilter related, but I happen to recall having a similar issue with ffdshow which could be solved by going to the ffdshow options, then "Subtitles" and then disable everything under "Subtitle sources", in particular "Embedded subtitles". If that doesn't work you should try asking in the XySubFilter thread.
That worked! So simple yet I overlooked it. Especially so as I had to do a similar thing to disable ffdshow interlacing (set it to "Bypass")

Unfortunately I have hosed ffdshow by updating to the latest 'tryout' figuring it might help. Going back to the previous build left me with issues concerning frame rate (30fps produces a hung image yet status (and audio) tells me it's playing..oh well..something to do :( Thanks again :)

Edit: Back in the saddle..yippie!

[flashmozzg]

to madshi: another parallel random number implementation. I know, you've already found one and I don't know how this will perform or if it evenapplicable, just posting it here for info.
https://pbs.twimg.com/media/BjRNi8nCQAASTcj.jpg:large

[Shiandow]

https://pbs.twimg.com/media/BjRNi8nCQAASTcj.jpg:large

That doesn't look like it will give a uniform random number at all, especially when s_uRandSeed is small, and when s_uRandSeed is too large s_uRand won't have enough significant bits. As far as I know the random number generator that MadVR uses is faster and more reliable.

[skypichat]

Hello, I have a problem with the full screen mode of madvr.
The default settings deactivate AERO in Windows 7 and my TV say that:
http://www.meedios.com/EN/forum/download/file.php?id=16014&mode=view
If I deactivate these settings in madVR, I don't get the problem.
http://www.meedios.com/EN/forum/download/file.php?id=16017&mode=view
http://www.meedios.com/EN/forum/download/file.php?id=16021&mode=view
I use madVR plugin of MeediOS.
Thank you for this best player. I use it every day

[kasper93]

Please use thumbnails.

[e-t172]

So, regarding my frame drop issues, I have grave news.

I swapped my PSU/motherboard/CPU/RAM and that made no difference. A friend just told me that he got himself a shiny new 780 Ti (so same GPU as mine but the card is from a different manufacturer) so I jumped on the opportunity and asked him to test for this problem. Turns out he had just installed Windows 8 so he had a clean, fresh system to do the testing.

He has *exactly* the same issue, namely frame drops every 10-30 minutes or so (with latest NVidia drivers, madVR defaults except smooth motion on).

This tends to indicate that every single 780 Ti out there has this frame drop problem. IF YOU ARE USING A 780 TI WITH MADVR, PLEASE MAKE YOURSELF KNOWN. Your input would be highly valuable.

This clearly smells like a pure driver issue that has not been fixed by NVidia yet and is affecting most if not all 780 Ti users. Tomorrow I'm going to dig in further and test every single NVidia driver version since the 780 Ti came out (331.93). I highly doubt this will reveal anything since I've already thoroughly tested 321.10 and it has the exact same issue, but one never knows. Does anyone know what would be the most efficient way of reporting this to NVidia?

In addition, it would be interesting if people with closely related GPUs (such as 780, 770, Titan etc.) could comment on whether they are able to achieve long zero-frame-drop playback (as shown in the stats) with madVR and smooth motion on.

[scollaco]

I couldn't find the answer to this anywhere...does Nvidia's edge enhancement feature work with madVR. From what I did as a quick test it didn't. Once I switched to EVR it did. Is this a known thing. Sorry if the question is silly.

[nevcairiel]

I couldn't find the answer to this anywhere...does Nvidia's edge enhancement feature work with madVR. From what I did as a quick test it didn't. Once I switched to EVR it did. Is this a known thing. Sorry if the question is silly.

It does not. madVR intentionally by-passes all GPU processing algorithms.

[Asmodian]

In addition, it would be interesting if people with closely related GPUs (such as 780, 770, Titan etc.) could comment on whether they are able to achieve long zero-frame-drop playback (as shown in the stats) with madVR and smooth motion on.

I will need to retest with the newest madVR but I have used madVR + smooth motion on my Titan for nearly a year now. I usually see 0 frame drops and repeats even after a long movie.

However, I have never been able to use smooth motion + fade detecting de-banding without reported frame drops. I always use debanding with both set to low to avoid fade detection.

[huhn]

Hello, I have a problem with the full screen mode of madvr.
The default settings deactivate AERO in Windows 7 and my TV say that:
If I deactivate these settings in madVR, I don't get the problem.

I use madVR plugin of MeediOS.
Thank you for this best player. I use it every day
madvr doesn't support 3d yet. and if i see this right your tv doesn't support 3d at 60 hz.
The default settings deactivate AERO in Windows 7 and my TV say

madvr doesn't disable aero with default settings.

[sexus]

I will need to retest with the newest madVR but I have used madVR + smooth motion on my Titan for nearly a year now. I usually see 0 frame drops and repeats even after a long movie.

However, I have never been able to use smooth motion + fade detecting de-banding without reported frame drops. I always use debanding with both set to low to avoid fade detection.

proud titan owner here, mind you its vbios modded aka boost 2.0 crap disabled , at a healthy boost setting of 1175 mhz on the gpu and a 3500 mhz on the ram at a power boost of 1.21v, with performance set to adaptive in the driver profile so it dont have no dreaded sudden performance drops like the horrible boost 2.0 does wich messed up fps like all hell

so ive noticed dropped frames or stuttering while using NNEDI3 from 16 up to 64 for chroma upscaling but 0 frame drops or stuttering with jinc3ar for chroma upscaling and NNEDI3 image doubling set to 64 for luma doubling and quadruppling and 16 for chroma doubling and quadruppling all set to always enabled , with Error diffusion option 1 and debanding at low and fade in fade out at high and the source was a 1080p/30fps video but does the same on lesser content aka 720p/30 and so on , mind you this

was on the older working opencl drivers aka 314.22 , as the current ones dont support NNEDI3 at all and logically not image doubling as well, hope we can see some improvement in that opencl performance department with fixed drivers , still no news i suppose

[Mangix]

Anyone know what API madVR uses to increase the clock resolution? That is, decrease the timer interval from 1.000 ms to 0.500 ms.

[Asmodian]

I will need to retest with the newest madVR but I have used madVR + smooth motion on my Titan for nearly a year now. I usually see 0 frame drops and repeats even after a long movie.

However, I have never been able to use smooth motion + fade detecting de-banding without reported frame drops. I always use debanding with both set to low to avoid fade detection.

I should have tested again before posting, I see no dropped frames even with smooth motion and debanding with fade detection.

edit: madVR v87.7, NNEDI3 64 neuron chroma, Jinc3 AR image, no doubling, smooth motion on, debanding low/high, ED2, no trade quality for performance. 1080p24 (bluray) -> 2560x1440 @ 75 Hz. Nvidia drivers 327.23

[leeperry]

Anyone know what API madVR uses to increase the clock resolution?
NtSetTimerResolution() (http://forum.doom9.org/showpost.php?p=1584898&postcount=13515)

[kasper93]

Any news from AMD guy or that was fake hope that someone really bother to help with our issue?

[PetitDragon]

http://www.avsforum.com/t/1477339/so-youve-built-your-htpc-now-what-is-next-how-to-get-the-ultimate-picture-and-sound-quality-from-your-htpc-madvr-svp-xbmc-mediabrowser-jriver/450#post_24514253

[e-t172]

I should have tested again before posting, I see no dropped frames even with smooth motion and debanding with fade detection.

edit: madVR v87.7, NNEDI3 64 neuron chroma, Jinc3 AR image, no doubling, smooth motion on, debanding low/high, ED2, no trade quality for performance. 1080p24 (bluray) -> 2560x1440 @ 75 Hz. Nvidia drivers 327.23

Thanks. If that's not too much to ask, would you mind testing for frame drops/presentation glitches for 1 hour or so but with madVR defaults (except smooth motion on), and using 60Hz instead of 75? Just trying to make sure our results are comparable. If you don't want to have to reconfigure afterwards you can save your madVR settings from the registry to a .reg file and then restore it afterwards.

[nevcairiel]

In addition, it would be interesting if people with closely related GPUs (such as 780, 770, Titan etc.) could comment on whether they are able to achieve long zero-frame-drop playback (as shown in the stats) with madVR and smooth motion on.

I can play full movies without frame drops on my 780 with smooth motion on, otherwise pretty conservative settings.
It shows an occasional glitch in the OSD (maybe 12 over a whole movie), but zero drops.

One thing i noticed when looking at the OSD for a while, the GPU seems to clock up and down a lot, as the presentation times vary quite a bit. For a while the max will be at 10ms and the average slowly going down, then for a while the max is 23ms and the average climbing until its nearly 16ms, then it clocks up again and average time goes back down.

[e-t172]

As promised I am currently sorting through all NVidia driver versions and it seems I've just hit the jackpot: after playing 1h20 of video with drivers 331.93 I am unable to reproduce the problem. Zero frame drops and zero glitches the entire time. Yay!

This result is quite unexpected because 321.10 did have issues. Maybe 321.10 has a separate bug (or maybe it doesn't actually work reliably with the 780 Ti).

Anyway, I will now be testing driver versions above 331.93 in an attempt to accurately identify the driver that introduced the regression.

Interestingly, even with 331.93 the spikes in XPerf are still there. That might mean that they are actually unrelated to the problem at hand. Or maybe I'm just bad at interpreting these graphs.

nevcairiel: what is your driver version?

[leeperry]

http://www.avsforum.com/t/1477339/so-youve-built-your-htpc-now-what-is-next-how-to-get-the-ultimate-picture-and-sound-quality-from-your-htpc-madvr-svp-xbmc-mediabrowser-jriver/450#post_24514253
Ah, you nail it! Everyone kept saying that there was no real-world diff between PCI-E 2.0 & 3.0 but I went for the latter anyway hoping that mVR's processing would be faster, and it is(dual-core Haswell/H87) :D

[nevcairiel]

nevcairiel: What is your driver version?

335.23

[huhn]

i have very low numbers with 3.0 so i'm not 100 % sure about this.
i test this with pcie 1-3 later but with render times and throughput.
not sure a low throughput adds a lot of rendertimes.

[e-t172]

As promised I am currently sorting through all NVidia driver versions and it seems I've just hit the jackpot: after playing 1h20 of video with drivers 331.93 I am unable to reproduce the problem. Zero frame drops and zero glitches the entire time. Yay!

Sigh...

Turns out I spoke too fast. I was able to get 2 hours straight of playback with zero frame drops/glitches, but after that I got two of them spaced ~20 minutes apart. It still seems like 331.93 is vastly better than the more recent drivers in that regard (e.g. it's only one frame each time and no glitches), but it's not 100% perfect as I let on in my previous message. I asked my friend with the 780 Ti to test 331.93 as well and he also got frame drops.

This means that in order to certify a working configuration I need to test for 2+ hours in order to be sure. Unfortunately my time isn't infinite so I think I'll just give up and wait for NVidia to fix their drivers. Right now I'm in a last-hope attempt to play with the madVR presentation, queue and flush settings to try to shake out these lone frame drops on 331.93.

[huhn]

put your card in highes powerstate and try it this way.

[e-t172]

Already tried that. Several times. Believe me, I've used every known trick in the book to try to shake out this bug.

(though it is true that I didn't try to play with the power state on 331.93, I'll try that next)

[daglax]

I noticed a strange behavior in madvr running an avisynth framedoubling script.
The avisynth-script doubles the source framerate, let's say to around 48 fps, but if i hit ctrl + j madvr still says, that the video frame rate is 23.976 fps.
Although avisynth is before madvr in the filter row, this is still happening.
And this leads to the problem, that madvr chooses the wrong profile and doesn't activate smooth motion or the right display clock.

Is there a way to get that right?

[seiyafan]

As promised I am currently sorting through all NVidia driver versions and it seems I've just hit the jackpot: after playing 1h20 of video with drivers 331.93 I am unable to reproduce the problem. Zero frame drops and zero glitches the entire time. Yay!


Have you tried 327.23? Can you get NNEDI3 to work with your version?

[e-t172]

To be honest I don't really care about NNEDI3 or other OpenCL-based madVR features at this point. I'm only using "traditional" madVR features such as 3DLUTs, classic upscaling, smooth motion, and the (theoretical in my case) ability to achieve glitch-free silky-smooth playback. I want to fix that first and then maybe look into these shiny new features.

If using 331.93 with all power management disabled and the GPU forced in P0 still doesn't work (and I don't think it will), I guess I could try 327.23 with a modded INF to make it work with the 780 Ti. However, considering that neither 331.93 nor 321.10 are able to provide flawless playback on my system (though 331.93 does come quite close), I would highly surprised if it made any difference.

[huhn]

I noticed a strange behavior in madvr running an avisynth framedoubling script.
The avisynth-script doubles the source framerate, let's say to around 48 fps, but if i hit ctrl + j madvr still says, that the video frame rate is 23.976 fps.
Although avisynth is before madvr in the filter row, this is still happening.
And this leads to the problem, that madvr chooses the wrong profile and doesn't activate smooth motion or the right display clock.

Is there a way to get that right?

try to add this in the file name [fps=48]. fps are reported to madvr and your script still reports 23.

[daglax]

try to add this in the file name [fps=48]. fps are reported to madvr and your script still reports 23.

Yeah, you're right, but only if the video is renamed before playing it.
That's a workaround, but not a real solution.
The solution, that comes to my mind, is that you could rename a video file on the fly with Avisynth, but i don't think this is possible.

Why is madvr not reading from the output instead of reading from the source filter?

[e-t172]

If using 331.93 with all power management disabled and the GPU forced in P0 still doesn't work (and I don't think it will), I guess I could try 327.23 with a modded INF to make it work with the 780 Ti. However, considering that neither 331.93 nor 321.10 are able to provide flawless playback on my system (though 331.93 does come quite close), I would highly surprised if it made any difference.

So... disabling power management made no difference with 331.93, and I'm unable to install 327.23 with a 780 Ti even with a modded INF (http://forum.doom9.org/showthread.php?p=1672170#post1672170).

I'm afraid I've reached a dead end for good this time. Anyone know of an effective way of reporting this to NVidia?

[kasper93]

Ah, you nail it! Everyone kept saying that there was no real-world diff between PCI-E 2.0 & 3.0


But it shouldn't. They are doing something wrong. It seems like they copyback to CPU RAM which is just insane. And this would explain why it's so slow on my HD5870. Generally knowing AMD I don't have big hopes that they do anything about it.

[huhn]

But it shouldn't. They are doing something wrong. It seems like they copyback to CPU RAM which is just insane. And this would explain why it's so slow on my HD5870. Generally knowing AMD I don't have big hopes that they do anything about it.

why should they change it? the number of programs using this is very low and it's currently working as intended.
so yeah i think so too.

[6233638]

One thing i noticed when looking at the OSD for a while, the GPU seems to clock up and down a lot, as the presentation times vary quite a bit.You could try using CUVID to force it into the high power state, or Nvidia Inspector's "multi display power saver" to force it into a low/medium power state and see if that helps.

When the GPU load falls into a place that it will change power states during playback, that is usually not seamless and you will have presentation glitches/dropped frames.

Ah, you nail it! Everyone kept saying that there was no real-world diff between PCI-E 2.0 & 3.0 but I went for the latter anyway hoping that mVR's processing would be faster, and it is(dual-core Haswell/H87) :DThere generally is no difference at all in the tests most sites perform. (e.g. gaming)

[leeperry]

IIRC this is the article (http://www.behardware.com/articles/850-2/pci-express-3-0-impact-on-performance.html) that led me to believe that DXVA would play nicer in mVR over 3.0:
going from the GPU to CPU there’s a nice 77% increase in bandwidth between PCI Express 3.0 x16 and 2.0 x16 (compared to 89% between 2.0 x16 and 1.0 x16)
the OpenCL applications that we were able to test are for now far from being limited by memory bandwidth. In already offering almost 7 GB/s, PCI Express 2.0 x16 covers most usages, even if certain very specific pro applications will certainly be able to take advantage of the theoretical gains.

[kiccolsd]

Those version numbers don't exist.

Sorry I was talking about v0.87.4 vs v0.87.7

[seiyafan]

IF YOU ARE USING A 780 TI WITH MADVR, PLEASE MAKE YOURSELF KNOWN. Your input would be highly valuable.

My 780 runs fine without frame drops.

[StinDaWg]

To be honest I don't really care about NNEDI3 or other OpenCL-based madVR features at this point. I'm only using "traditional" madVR features such as 3DLUTs, classic upscaling, smooth motion, and the (theoretical in my case) ability to achieve glitch-free silky-smooth playback. I want to fix that first and then maybe look into these shiny new features.
If you watch a lot of 720p content on a 1080p screen you really should be using nnedi. Upscaling quality is way better than Jinc. It's sharper than Lanczos with basically no ringing or aliasing. It's like using a sharpening filter with none of the side effects. The difference is pretty amazing to be honest. Even if your card can only get to Bicubic 75 for chroma and luma doubling at 16 neurons, you are 95% of the way there IMO.

[Qaq]

Some madVR users don't really care of its PQ and use madVR only for features.

[StinDaWg]

Some madVR users don't really care of its PQ and use madVR only for features.
Which of its "features" are not directly related to picture quality?

[Asmodian]

Thanks. If that's not too much to ask, would you mind testing for frame drops/presentation glitches for 1 hour or so but with madVR defaults (except smooth motion on), and using 60Hz instead of 75? Just trying to make sure our results are comparable. If you don't want to have to reconfigure afterwards you can save your madVR settings from the registry to a .reg file and then restore it afterwards.

I gave this a shot, 1h 20min of anime bluray (24p), 0-0-0 drop-delay-glitch. I reset the stats after switching to full screen because with the default no playback delay some frames are dropped on open and when switching to full screen. Gets up to ~11ms rendering times due to throttling down.

Completely default madVR settings with madVR smooth motion enabled in Zoom Player. I also set my display to 60 Hz.

I don't see any odd usage spikes in GPU, CPU, or IO meters. SLI off.

[Q-the-STORM]

Which of its "features" are not directly related to picture quality?

display mode switching, IVTC, smooth motion and Exclusive Mode... that's pretty much it...

[Asmodian]

Those all improve motion quality instead of picture quality which seems like a similar goal to me. :)

@madshi
Is this something easy to implement and actually worth it?

I like profiles a lot but I would like the ability to multiply in the if statement or maybe a single new variable instead, pixels-per-second.
This would help a lot with lower resolution interlaced material, I want to set higher neuron counts for 480p30 than 480p60 but I also don't want very complex or numerous if statements. I could use the pixels-per-second instead of width, height, and deintfps so 480p60 would automatically use my 720p30 profile and 720p60 would automatically use my 1080p30 profile. :D

[StinDaWg]

display mode switching, IVTC, smooth motion and Exclusive Mode... that's pretty much it...
First and last aren't exclusive to madVR. Pretty much any decent video player will have them, including MPC-HC out of the box. Smooth motion does more harm to the picture than good (IMO). I use IVTC all the time, very handy. Anyways...