madVR - high quality video renderer (GPU assisted)

[huhn]

Stereodude: How do you plan on using nvidia's ivtc? Program, renderer, decoder, etc.

when you run madvr in video mode the deinterlacer from the gpu is used and there is an option for "use inverse telecine" the nvidia type does an good work with "field matching" but no frames are dropped/decimated and it's 2:3 (output is 60 fps) not 1:1 so it judder a lot and the picture quality is lower with nvidia compared to madvr ivtc maybe nvidia uses very good bob? i have not tested 30 fps deinterlacing from cuda maybe that has the same quality as madvr ivtc but with a repeated frame.

this deinterlacer should be the same deinterlacing as in CUDA.

[DragonQ]

i have not tested 30 fps deinterlacing from cuda maybe that has the same quality as madvr ivtc but with a repeated frame.
Yes, it's just weave with each frame repeated.

[turbojet]

Doing what huhn mentioned is not a good result with nvidia, combing artifacts quite often and like dragonq mentioned every frame is repeated which at times looks like a bunch of blended frames at 60fps. The only thing enabling nvidia ivtc does it looks like is reduce the combing artifacts a bit but still unwatchable imo.

[huhn]

Doing what huhn mentioned is not a good result with nvidia, combing artifacts quite often and like dragonq mentioned every frame is repeated which at times looks like a bunch of blended frames at 60fps. The only thing enabling nvidia ivtc does it looks like is reduce the combing artifacts a bit but still unwatchable imo.

are judging this with your 250 gts? if yes new cards like the 760 gtx does an very good job on the samples you posted above. only the judder is very annoying but there are no miss matched or at least i didn't see them at all. it's not like nvidia hasn't changed the past years.

[DragonQ]

Doing what huhn mentioned is not a good result with nvidia, combing artifacts quite often and like dragonq mentioned every frame is repeated which at times looks like a bunch of blended frames at 60fps. The only thing enabling nvidia ivtc does it looks like is reduce the combing artifacts a bit but still unwatchable imo.
25 fps doubled to 50 fps looks the same as 50 fps to me. Maybe because plasma already sub-samples loads anyway. That's with a GT 430, but I also don't notice a difference using my GTS 250 on a PC monitor (60 Hz with Smooth Motion).

This has nothing to do with the IVTC setting though; like you say, enabling it doesn't prevent 25/30p being presented as 50/60p if the stream is marked as interlaced.

[00pontiac]

(snip)
all my bluray rips are from my own real blurays and all ripped with my one default dvdfab profile, that is to say they're all pretty much identical. some videos are starting paused and some don't. it's always the same videos that start paused and the same ones that don't. as an example Game of Thrones season 2 episode 1 starts normally but season 2 episode 9 starts paused. other shows and movies do this too but what they are is probly insignificant.
(snip)


The bug tracker is for issues which are likely to be bugs in madVR, and which occur for more than just one user. If you're not sure, it makes sense to ask here first. In your specific case we should first check a few things, because I don't remember anyone else reporting your issue recently. So it could be a misconfiguration or some other filter making problems.

Could the problem have to do with the subtitles somehow? E.g. playback starting if there are no subtitles and not starting when there are some?

This problem is probably related to the "delay playback start until queues are full" option in the madVR settings.


hey! that's it! except the other way around. the videos that are pausing at startup are the videos without any subtitles. I'm using DirectVobSub as my Subtitle Renderer, I've tried with FFDShow too but no difference. I've tried disabling FFDShow subtitle rendering in Zoomplayer for both when VobSub is preferred and when no subtitle files/streams are found. not sure where to go from here....

I have an ongoing post at the zoomplayer forums with Ehat on this and another problem too, and I just posted there with this info, maybe he will come up with something. I'm a bit at a loss.

Thanks!

[huhn]

try a faster subtitle renderer like xy vobsub. the normal direct vob sub is very slow is bad for quality because is renders in rgb.
is there no build in overlay subtitle renderer in zoomplayer like in mpc hc?

[turbojet]

That was on a gtx 250 on a 650 the combing is removed but during playback it still looks like blended frames, maybe 7xx series is better? In both of my cases evr cp's deinterlacing was much more pleasing but not nearly as much as madvr's ivtc >95% of the time. A few tweaks by madshi could make madvr's ivtc the obvious choice for everything.

EDIT: MadVR see's 2:2 cadence on the whole episode this clip (http://www.mediafire.com/download/5l6llo814918rff/wrong_cadence_banding.mpg) is from, it should be 3:2. There's also a lot of banding in the last scene that might be helpful.

[DragonQ]

That was on a gtx 250 on a 650 the combing is removed but during playback it still looks like blended frames, maybe 7xx series is better? In both of my cases evr cp's deinterlacing was much more pleasing but not nearly as much as madvr's ivtc >95% of the time. A few tweaks by madshi could make madvr's ivtc the obvious choice for everything.
I'm confused. "EVR's deinterlacing" is the GPU's deinterlacing. That's what we're talking about, isn't it? If deinterlacing is enabled in MadVR it'll also use the GPU's algorithm(s), unless you force film mode, in which case MadVR will try to work out the cadence and recover the frames appropriately.

The nVidia cards you mention can do 25i/30i to 50p/60p perfectly well and will detect 2:2 cadence correctly too (but still output at 50p/60p). I have no idea whether they can do 3:2 pulldown well though; I'm sure MadVR is better at that.

EDIT: MadVR see's 2:2 cadence on the whole episode this clip (http://www.mediafire.com/download/5l6llo814918rff/wrong_cadence_banding.mpg) is from, it should be 3:2. There's also a lot of banding in the last scene that might be helpful.
Looks like 2:2 cadence to me (30p). Disabling deinterlacing completely results in 29.97 fps (according to MadVR) and stepping through each frame shows they are complete with no combing or repeating. Why do you think it should be 3:2 cadence (24p)?

[PetitDragon]

EDIT: MadVR see's 2:2 cadence on the whole episode this clip (http://www.mediafire.com/download/5l6llo814918rff/wrong_cadence_banding.mpg) is from, it should be 3:2. There's also a lot of banding in the last scene that might be helpful.

Yep, this clip should be 3:2 for sure.;)

[Mirage!]

I found a small bug in the latest test madVR (possibly even in the one on the first page). The checkbox "enable smooth motion frame rate conversion" is cleared every time I restart the media player. I use Windows 8.1, and it could be caused by the file system protection in Program Files (it can't be turned off easily in Win8). I don't quite understand how the settings are saved. There are duplicate settings in settings.bin, and in the registry.

[DarkSpace]

EDIT: MadVR see's 2:2 cadence on the whole episode this clip (http://www.mediafire.com/download/5l6llo814918rff/wrong_cadence_banding.mpg) is from, it should be 3:2. There's also a lot of banding in the last scene that might be helpful.
Perhaps the whole episode is different, but this clip detects as 3:2 throughout the whole video on the latest debanding test build.

[AlexCeed]

OIn that case it could be madVR's fault, or it could be the media player's fault. Which media player are you using? And does the problem also occur if you e.g. use EVR in fullscreen exclusive mode?

I'm using MPC-HC nightly build and this problem occurs only in madVR. EVR works OK.

[Stereodude]

when you run madvr in video mode the deinterlacer from the gpu is used and there is an option for "use inverse telecine" the nvidia type does an good work with "field matching" but no frames are dropped/decimated and it's 2:3 (output is 60 fps) not 1:1 so it judder a lot and the picture quality is lower with nvidia compared to madvr ivtc maybe nvidia uses very good bob? i have not tested 30 fps deinterlacing from cuda maybe that has the same quality as madvr ivtc but with a repeated frame.

this deinterlacer should be the same deinterlacing as in CUDA.I'm not sure what you're talking about. Nvidia had pretty much nailed deinterlacing and IVTC handling going back to the GT 8600 or earlier through subsequent driver revisions. Didn't you ever see the testing over the years with the Silicon Optix HQV Benchmark DVD? link (http://hothardware.com/Reviews/NVIDIA-GeForce-8800-GT--G92-Takes-Flight/?page=11)

You're telling me in the past 5 years they've totally screwed it all up and can't do it right anymore? This seems to say otherwise (http://www.anandtech.com/show/3973/nvidias-geforce-gt-430/4).

[mindbomb]

amd has recently announced their version of smooth motion called fluid motion. I think this opens up the possibility for hardware ivtc to not have judder. But yea, up until now, judder was still a problem.
http://images.anandtech.com/doci/7677/36%20-%20Fluid%20Motion%20Video.jpg

[Stereodude]

Judder is an artifact of the frame rate differences. There's no way to get rid of it on film based material unless the display supports frame rates that are a multiple of 24 or you interpolate new frames. The presence of judder hardly indicates that the hardware processing is low quality. In various testing in the past I recall both AMD and Nvidia cards changing the frame rate passed to the EVR to a multiple of 24 (23.976) on what started as i60 content in a file.

[e-t172]

amd has recently announced their version of smooth motion called fluid motion. I think this opens up the possibility for hardware ivtc to not have judder. But yea, up until now, judder was still a problem.
http://images.anandtech.com/doci/7677/36%20-%20Fluid%20Motion%20Video.jpg

What's in this slide is frame interpolation, not frame blending. It is not comparable to madVR SmoothMotion.

[turbojet]

DragonQ: I was testing EVR in MPC-BE vs. madvr in Potplayer. Potplayer has it's own deinterlace settings and when set to disabled it disables the gpu deinterlacing in evr/madvr. In order for madvr ivtc to work correctly it needs to be disabled. Comparing both renderers in mpc-be with a hard telecined film source, they both deinterlace but madvr leaves a lot of judder, unwatchable imo, while evr is smooth, display is at 60hz tested on nvidia 250 and 650, nvidia ivtc didn't seem to have any effect. With a 30 fps video source they both play fine like you mentioned.

The wrong_cadence file is filmed and while it appears to have no dupe frames when frame stepping. There are 2 of 5 nearly identical frames before deinterlacing and it ivtc's well, http://www.mediafire.com/download/tyqsulciimyviq5/wrong_cadence_banding_ivtcd.mkv

Darkspace: both clip and episode show 2:2 here with mpc-be and potplayer, what player are you using and what madvr deinterlace settings?

Stereodude: Nvidia ivtc may work very well with evr but not with madvr on a 250 and 650 ime.

The cadence_extreme file from earlier got flagged by mediafire for copyright and was blocked. I shortened it up and it along with the other madvr ivtc problem files can be found at https://www.mediafire.com/folder/3sjqpvk25lzq6/madvr%20ivtc

[Asmodian]

I found a small bug in the latest test madVR (possibly even in the one on the first page). The checkbox "enable smooth motion frame rate conversion" is cleared every time I restart the media player. I use Windows 8.1, and it could be caused by the file system protection in Program Files (it can't be turned off easily in Win8). I don't quite understand how the settings are saved. There are duplicate settings in settings.bin, and in the registry.

Are you using Zoom Player?
If so it actually has its own setting for smooth motion that overwrites madVR's.
Options->Playback->Video->Enable MadVR's smooth motion

[DarkSpace]

Darkspace: both clip and episode show 2:2 here with mpc-be and potplayer, what player are you using and what madvr deinterlace settings?
My player is the latest stable MPC-HC, my madVR deinterlacing settings are:
automatically activate deinterlacing when needed (if in doubt, deactivate)
disable automatic source type detection (force film mode)
not only look at pixels in the frame center

[dansrfe]

I'm not sure if this is relevant but why is D3D11 not used? Does it not offer any performance improvements over D3D9 for what madVR uses?

[turbojet]

DarkSpace: I can duplicate your results. If lav video decoder is used to decode the correct cadence 3:2 is detected, if ffdshow, internal mpc-be or potplayer decoders are used it detects 2:2. Potplayer also must have filter control>condition disabled, even if no processing is enabled, to work correctly with this and the evr/madvr deinterlacing. This looks to be reducing the cadence breaks quite a bit, what could cause the difference between lav video and the other decoders?

[jkauff]

I found a small bug in the latest test madVR (possibly even in the one on the first page). The checkbox "enable smooth motion frame rate conversion" is cleared every time I restart the media player. I use Windows 8.1, and it could be caused by the file system protection in Program Files (it can't be turned off easily in Win8). I don't quite understand how the settings are saved. There are duplicate settings in settings.bin, and in the registry.
You might try running your player using the "Run as Administrator" right-click option before you change any settings.

[Mirage!]

Are you using Zoom Player?
If so it actually has its own setting for smooth motion that overwrites madVR's.
Options->Playback->Video->Enable MadVR's smooth motion
Superb, thanks, mate. I'm using Zoom Player, and enabling this setting fixed it. I'm not sure why they chose to include this setting (and not other).

[DarkSpace]

DarkSpace: I can duplicate your results. If lav video decoder is used to decode the correct cadence 3:2 is detected, if ffdshow, internal mpc-be or potplayer decoders are used it detects 2:2.
Indeed, I am using the latest stable LAV. My guess here is that LAV either adds or doesn't add some flags that lead madVR to detect a 3:2 cadence.
I have no experience in playing back soft-telecined material, but could it be that the file is soft-telecined (at least parts of it) and LAV outputs it to a constant 30 fps and tells madVR which fields have been duplicated, and how?

[madshi]

hey! that's it! except the other way around. the videos that are pausing at startup are the videos without any subtitles. I'm using DirectVobSub as my Subtitle Renderer, I've tried with FFDShow too but no difference. I've tried disabling FFDShow subtitle rendering in Zoomplayer for both when VobSub is preferred and when no subtitle files/streams are found. not sure where to go from here....
Not sure. Have you tried xy-vsfilter? I'm not sure why the problem occurs. Try disabling/blocking the DirectVobSub auto-loading filter. Does that "fix" the problem?

The nVidia cards you mention can do 25i/30i to 50p/60p perfectly well and will detect 2:2 cadence correctly too (but still output at 50p/60p). I have no idea whether they can do 3:2 pulldown well though; I'm sure MadVR is better at that.
2:2 film mode vs video mode is pretty hard to get right and last time I checked NVidia fails here quite often. I've seen NVidia incorrectly handling film content as video content with several samples. I quite disagree with your statement of NVidia doing 25i/30i to 50p/60p "perfectly".

I'm using MPC-HC nightly build and this problem occurs only in madVR. EVR works OK.
Do you know a freeware game with which is able to run in fullscreen exclusive mode on a secondary display? That would be good for testing. In any case, please feel free to report this problem to the madVR bug tracker. I don't know when I'll find the time to look into this problem, though.

Nvidia had pretty much nailed deinterlacing and IVTC handling going back to the GT 8600 or earlier through subsequent driver revisions. Didn't you ever see the testing over the years with the Silicon Optix HQV Benchmark DVD?
What does the HQV Benchmark DVD have to do with anything? People put way too much stock into those test results. It contains just a bunch of test scenes. Real world material is often more difficult than those test scenes. Furthermore most of those test scenes are for video content. Only 2 of all those tests are targetted at film content.

NVidia does not have deinterlacing and IVTC "nailed" at all. It works reasonably well, but there's room for improvement. Of course the biggest problem is the lack of decimation, but we can't fault NVidia for that, it's a limitation of the DXVA deinterlacing API. But there are other problems, as well, e.g. reliable detection of PAL film vs video. Which btw the HQV Benchmark doesn't really stress test at all.

I'm not sure if this is relevant but why is D3D11 not used? Does it not offer any performance improvements over D3D9 for what madVR uses?
D3D11 does not support Vista and XP, both of which madVR still supports. Also DXVA is usually still based on D3D9 (although win7 now also supports hardware accelerated decoding with D3D11, I think).

DarkSpace: I can duplicate your results. If lav video decoder is used to decode the correct cadence 3:2 is detected, if ffdshow, internal mpc-be or potplayer decoders are used it detects 2:2.
That means that this content is soft-telecined and those decoders don't seem to properly forward the "repeat_first_field" flag to madVR. It's a bug in those decoders, not in madVR. No reason to use any of those decoders instead of LAV, anyway.

[DragonQ]

2:2 film mode vs video mode is pretty hard to get right and last time I checked NVidia fails here quite often. I've seen NVidia incorrectly handling film content as video content with several samples. I quite disagree with your statement of NVidia doing 25i/30i to 50p/60p "perfectly".
The phrase "perfectly well" is more akin to "adequately" than "perfectly". However, the only times I've seen irregularities it's been with mixed content (e.g. primarily interlaced but with progressive graphics in certain parts). From my testing, these issues are visible when using any deinterlacer (including those found on TVs) and are sometimes even problems at the encoder end, so I don't know of any way around this.

I don't think I've ever seen film content incorrectly treated as video content. What would that even look like? This has come up more than once in this thread and every time a 2:2 cadence sample is posted it gets weaved perfectly (not perfectly well :p) on my nVidia GPUs.

I do find it odd that 2:2 cadence is supposedly so difficult to detect though, particularly with 25i streams. Even high end TVs sometimes can't do it properly according to reviews.

[madshi]

Well, if film content is treated as video content, you don't see combing artifacts. Instead the image is just not as calm and high resolution as it would be if properly IVTCed. You might not notice the problem at real time watching speed, if you don't have a reference to compare to.

[DragonQ]

Well, if film content is treated as video content, you don't see combing artifacts. Instead the image is just not as calm and high resolution as it would be if properly IVTCed. You might not notice the problem at real time watching speed, if you don't have a reference to compare to.
Makes sense. Like I said though, I've never seen this. On the rare occasions where I've checked via frame-stepping, the video has always been weaved correctly. I wouldn't be surprised if it took a few frames to "lock on" to the 2:2 cadence though.

[madshi]

It probably works correctly most of the time. But there are scenes where it fails.

[Stereodude]

What does the HQV Benchmark DVD have to do with anything? People put way too much stock into those test results. It contains just a bunch of test scenes. Real world material is often more difficult than those test scenes. Furthermore most of those test scenes are for video content. Only 2 of all those tests are targetted at film content.

NVidia does not have deinterlacing and IVTC "nailed" at all. It works reasonably well, but there's room for improvement. Of course the biggest problem is the lack of decimation, but we can't fault NVidia for that, it's a limitation of the DXVA deinterlacing API. But there are other problems, as well, e.g. reliable detection of PAL film vs video. Which btw the HQV Benchmark doesn't really stress test at all.He said it didn't work and suggested they were using a Bob. I was simply pointing to some objective 3rd party measurements / test to show that it does work (in at least those test cases). That seemed better than saying, "It seems to work pretty well for me."

Where I live I'm pretty much never going to encounter PAL film content. It's either going to be film with a telecine to make it i60 which needs an IVTC, or true interlaced video content. IMHO, what would be nice would be user side options allowing some of the cadence patterns to be blocked in the IVTC routine, and some way to configure madVR so that it will attempt an IVTC and if the IVTC fails it will pass the footage to the video card driver for "video" deinterlacing. Currently it seems I have to disable automatic source type detection and force film for the IVTC routines in madVR to work, and if I play back true interlaced video content it's not going to be properly handled until I enable automatic source type detection again.

[flashmozzg]

D3D11 does not support Vista and XP, both of which madVR still supports. Also DXVA is usually still based on D3D9 (although win7 now also supports hardware accelerated decoding with D3D11, I think).

It supports Vista (or in other words Vista supports it). Only xp is left out. But xp will soon become not supported even by microsoft.

[madshi]

IMHO, what would be nice would be user side options allowing some of the cadence patterns to be blocked in the IVTC routine, and some way to configure madVR so that it will attempt an IVTC and if the IVTC fails it will pass the footage to the video card driver for "video" deinterlacing. Currently it seems I have to disable automatic source type detection and force film for the IVTC routines in madVR to work, and if I play back true interlaced video content it's not going to be properly handled until I enable automatic source type detection again.
The current madVR deinterlacing solution is just a first step. More steps to come at some point in the future. But probably not soon.

It supports Vista. Only xp is left out.
You're right. It's the 3D features of D3D11.1 which are not supported on Vista.

[nevcairiel]

You're right. It's the 3D features of D3D11.1 which are not supported on Vista.

Those are not even in 7, aren't they?

[madshi]

Oh yes, I think you're right! Seems my mind is playing tricks on me today. Need some sleep... :)

[DragonQ]

Those are not even in 7, aren't they?
.....
The Platform Update for Windows 7 includes a limited set of features from Direct3D 11.1, though components that depend on WDDM 1.2 - such as feature level 11_1 and its related APIs, or quad buffering for stereoscopic rendering - are not present.

[turbojet]

The 2:2, 3:2 file is indeed soft telecine, 96.55% film according to dgindex.

Where I live I'm pretty much never going to encounter PAL film content. It's either going to be film with a telecine to make it i60 which needs an IVTC, or true interlaced video content. IMHO, what would be nice would be user side options allowing some of the cadence patterns to be blocked in the IVTC routine, and some way to configure madVR so that it will attempt an IVTC and if the IVTC fails it will pass the footage to the video card driver for "video" deinterlacing. Currently it seems I have to disable automatic source type detection and force film for the IVTC routines in madVR to work, and if I play back true interlaced video content it's not going to be properly handled until I enable automatic source type detection again.

If madvr looked for a 3:2 pattern and if not found switched to deinterlacing it would work for almost everything out there. While I haven't done much watching with the 'fixed lav video' setup, doing this would solve almost all the problems I currently have. It keeps trying to find a cadence and can't. cadence.mpg is fixed with lav video. cadence_extreme.mpg and combing.mpg should be fixed if searching only for 3:2 or deinterlace. The interlaced overlay 'bottom_combing.mpg' and motion_bottom.mpg problems would still exist but maybe something can be found in IT.dll source (which handles both great).

EDIT: Would the repeat_first_flag or something else have an effect on hard telecined material? Switching to it reduce cadence breaks from about 1 per minute on average to 1 every 2.5-3 minutes.

[huhn]

Do you know a freeware game with which is able to run in fullscreen exclusive mode on a secondary display? That would be good for testing. In any case, please feel free to report this problem to the madVR bug tracker. I don't know when I'll find the time to look into this problem, though.

starcraft 2 is free to play (at least a part of it) and it has an option to set the default monitor, perfect for this test. but it's big...

https://us.battle.net/account/sc2/starter-edition/

[6233638]

D3D11 does not support Vista and XP, both of which madVR still supports. Also DXVA is usually still based on D3D9 (although win7 now also supports hardware accelerated decoding with D3D11, I think).I know that you want to support as many operating systems as possible, but Vista does have support for D3D11, and even Microsoft is dropping support for XP in April.

I'm not saying that you should rip out the existing D3D9 functionality, but if supporting D3D11 would improve madVR in a meaningful way (e.g. 10-bit output) it's something to consider.

[Ver Greeneyes]

I know that you want to support as many operating systems as possible, but Vista does have support for D3D11, and even Microsoft is dropping support for XP in April.
I'd say a bigger problem with Direct3D 11 is graphics card support - for instance, only the GeForce GT 430 and up support D3D11. The first GeForce 4xx cards became available in April 2010 (whereas Vista dates back to January 30, 2007). Mind you that's mostly a reason to not rip out Direct3D 9 support.

[flashmozzg]

I'd say a bigger problem with Direct3D 11 is graphics card support - for instance, only the GeForce GT 430 and up support D3D11. The first GeForce 4xx cards became available in April 2010 (whereas Vista dates back to January 30, 2007). Mind you that's mostly a reason to not rip out Direct3D 9 support.

Dx11 is backwards compatible with Dx10, though some dx11-specific features are not supported on Dx10 cards (like HW tessellation)

[Stereodude]

Should FSE have higher rendering time than Windowed mode? I just got a Radeon HD7790 using the 13.12 drivers under Windows 7 (32-bit) with MPC-HC 1.71 that's what I'm seeing. In FSE I can barely keep the rendering time under 1/60th of a second if I crank up the madVR settings with all the driver video "enhancements" disabled. In Windowed mode I can use the max madVR settings and enable all of the video enhancements (except edge-enhancement) and still have a few ms of rendering time to spare.

[cyberbeing]

@Stereodude

If you've set the number of video frames presented in advance too high, maybe. AMD cards behave better with low values like 6 to avoid DPC latency issues.

If you are using a refresh rate higher than your video, FSE is slightly more intensive since it presents a frame every vsync interval, while Windowed mode only presents a single time for each video frame.

If you've set Flush or Flush & Wait for any of the render steps, that will also increase render times and/or tendency for dropped frames quite a bit under heavy GPU load.

Otherwise, ensure you are not forcing Anti-Aliasing, Anisotropic Filtering, VSync, etc globally under your GPU's 3D settings. Any globally forced settings will be detrimental to madVR performance under FSE. GPU video "enhancement" settings should also be left disabled if you are using madVR.

[Stereodude]

If you've set the number of video frames presented in advance too high, maybe. AMD cards behave better with low values like 6 to avoid DPC latency issues.

If you are using a refresh rate higher than your video, FSE is slightly more intensive since it presents a frame every vsync interval, while Windowed mode only presents a single time for each video frame.

If you've set Flush or Flush & Wait for any of the render steps, that will also increase render times and/or tendency for dropped frames quite a bit under heavy GPU load.

Otherwise, ensure you are not forcing Anti-Aliasing, Anisotropic Filtering, VSync, etc globally under your GPU's 3D settings. Any globally forced settings will be detrimental to madVR performance under FSE.I pretty much haven't changed any of the default settings in madVR. I will take a look at some of your suggestions later. I did find that disabling Aero in full screen exclusive mode dropped the FSE rendering time several ms.

GPU video "enhancement" settings should also be left disabled if you are using madVR.Aren't these "enhancements" done ahead of the madVR render? If so, some of these features and madVR aren't mutually exclusive. Sure some of the features should be avoided like the plague like edge-enhancement and dynamic contrast, but the noise reduction features when used in moderation can be beneficial.

[madshi]

Would the repeat_first_flag or something else have an effect on hard telecined material? Switching to it reduce cadence breaks from about 1 per minute on average to 1 every 2.5-3 minutes.
The repeat_first_flag field only has an affect on soft-telecined content. If you see reduced cadence breaks with LAV decoder compared to the other decoders than that probably means that the content you're testing with is not purely hardware-telecined, but that it is probably a mixture of soft- and hard-telecine. Which is the case for many broadcasts and DVDs...

starcraft 2 is free to play (at least a part of it) and it has an option to set the default monitor, perfect for this test. but it's big...

https://us.battle.net/account/sc2/starter-edition/
Hmmmm... Does it actually switch the default monitor in the Windows settings? Meaning: Does the taskbar switch over to the other monitor while playing? Or does it only play the game on the other monitor while leaving the start menu on the original monitor?

Should FSE have higher rendering time than Windowed mode? I just got a Radeon HD7790 using the 13.12 drivers under Windows 7 (32-bit) with MPC-HC 1.71 that's what I'm seeing. In FSE I can barely keep the rendering time under 1/60th of a second if I crank up the madVR settings with all the driver video "enhancements" disabled. In Windowed mode I can use the max madVR settings and enable all of the video enhancements (except edge-enhancement) and still have a few ms of rendering time to spare.
FWIW, you shouldn't put too much weight into the rendering times. They might be accurate, or they might not. E.g. they're not accurate if you don't have a "flush & wait" of some kind set for the last rendering step. It's more important to check for dropped frames. In my experience windows mode drops frames earlier than FSE mode, if rendering times get near to the "max". That applies especially to "smooth motion FRC". With e.g. 24fps @60Hz playback with smooth motion FRC enabled, I've found that the rendering times in windowed mode must be smaller than 1/60s, while in exclusive mode they only need to be smaller than 1/24s, to achieve smooth playback. Anyway, as long as no frames are dropped with any specific settings, you're fine.

Most of the video driver "enhancement" algorithms only affect the image (and the rendering times) if you use either DXVA deinterlacing or DXVA scaling. Otherwise they should have no effect.

[Vyral]

@Stereodude
I've got a AMD HD 6800 series (13.9 drivers). The following settings works perfectly on madVR (less than 10 delayed frames on 2 hours video and no dropped frames) :
Render queue (CPU & GPU) : 12
Number of video frames presented in advance : 6
I disabled all "enhancements" together with IVTC processing in CCC and set the output to RGB 4:4:4 full range.
Moreover, HW acceleration increase the redering time, see if it's enabled in MPC-HC > LAV video settings

[huhn]

Hmmmm... Does it actually switch the default monitor in the Windows settings? Meaning: Does the taskbar switch over to the other monitor while playing? Or does it only play the game on the other monitor while leaving the start menu on the original monitor?
i was trying to find this out and realize it is not possible i mixed the game up or something else i'm sorry. i'm still confused about this...

[Stereodude]

FWIW, you shouldn't put too much weight into the rendering times. They might be accurate, or they might not. E.g. they're not accurate if you don't have a "flush & wait" of some kind set for the last rendering step. It's more important to check for dropped frames. In my experience windows mode drops frames earlier than FSE mode, if rendering times get near to the "max". That applies especially to "smooth motion FRC". With e.g. 24fps @60Hz playback with smooth motion FRC enabled, I've found that the rendering times in windowed mode must be smaller than 1/60s, while in exclusive mode they only need to be smaller than 1/24s, to achieve smooth playback. Anyway, as long as no frames are dropped with any specific settings, you're fine.Well, it was also dropping frames in FSE, but wasn't in Windowed. However, I was using DXVA native in LAV which end up reporting the clips as 59.94fps instead of 29.97 for 1080i60 files. Using DXVA copy back or software decoding the clip frame rate was reported as 29.97 and the dropped frames stopped. While still using DXVA native mode I was also able to reduce the reported rendering time and eliminate the dropped frames in FSE by disable desktop composition in full screen mode.

Most of the video driver "enhancement" algorithms only affect the image (and the rendering times) if you use either DXVA deinterlacing or DXVA scaling. Otherwise they should have no effect.That's good information to have. I don't mind using some of the noise reduction features with light settings. However, I didn't realize they weren't active if madVR was doing IVTC and the sophisticated scaling algorithms. Frankly, I'd rather have smooth motion than some light noise reduction.

@Stereodude
I've got a AMD HD 6800 series (13.9 drivers). The following settings works perfectly on madVR (less than 10 delayed frames on 2 hours video and no dropped frames) :
Render queue (CPU & GPU) : 12
Number of video frames presented in advance : 6
I disabled all "enhancements" together with IVTC processing in CCC and set the output to RGB 4:4:4 full range.
Moreover, HW acceleration increase the rendering time, see if it's enabled in MPC-HC > LAV video settings
I've been testing with the defaults of 16 & 8 respective for the queues. Frames in advance for FSE is 4 (default). I didn't notice HW acceleration in LAV increasing the rendering time / dropped frame count (aside from the DXVA native vs. DXVA copy back thing already mentioned), but I will do more testing today. I have been able to find settings that give me no dropped frames with all the features I wanted enabled and that was the important thing.

[madshi]

FWIW, the next madVR version will have a default number of pre-presented frames of 8 (instead of the current default value of 4).

[huhn]

found the real game with that option. it's world of tanks. it doesn't change the taskbar it's free to play and if i get it right it's direct x 9.