madVR - high quality video renderer (GPU assisted)

[70MM]

Still I find playing ripped BDs at 59.94Hz with smooth motion on doesnt look that great, there is too much blur going on with the blended frames.
I used it this way for a very long time, but finally went back to setting everything up to play in its native frame rate, 23.98 in > 23.98 out, no smooth motion.

However having a perfect clock run without a dropped or repeated frame is quiet hard to master....
Although I dont get many, I do get them....

Im waiting and hoping what madshi is working on in this area does come to fruition.

[AngelGraves13]

I hope that sometime in the future that NGU Chroma will only scale once with image doubling, like SuperXBR instead of scaling once and then being doubled.

[mrjones]

I am playing live TV through kodi/Vader and am getting buffering .
MADVR is reporting the clock deviation is at 68% !!!
Is this the source or a function of something else?

[QBhd]

I am playing live TV through kodi/Vader and am getting buffering .
MADVR is reporting the clock deviation is at 68% !!!
Is this the source or a function of something else?

That's normal... as soon as there is a glitch in the stream or buffering that "clock deviation" is meaningless. It is a calculated guess at the clock deviation, and the calculation goes through the roof if the stream gets interrupted.

QB

[Asmodian]

Interesting discussion on the topic. Yeah, that is my prime question: does smooth motion nullify accurate clock timings. Less blended frames is a good thing. I'm sure madshi will answer too.

I found a quote from madshi about it, I remember another one mentioning not syncing the start of the blending factors to v-sync but I couldn't find it.
The reason madVR measures the VSync position for blending factors does not have anything to do with tearing. Actually I could add any random (but constant!) offset to the measured VSync position, and it wouldn't cause motion smoothness to be harmed. E.g. if I added a 0.5 frame offset to the measured VSync position, this would cause the blending factors to change accordingly, and the net result would be an audio/video sync mismatch of 0.5 frames, but motion smoothness would still be the same. The reason madVR measured the VSync position is that only this way the blending factors adjust perfectly to the output refresh rate, so that I never ever have to drop or repeat a video frame.

It's complicated, but I hope my explanation is understandable?


I did some experimenting in Excel:
24/1.001 fps @ 59.900 Hz = 80.04% blended
24/1.001 fps @ 60/1.001 Hz = 40.12% blended
24/1.001 fps @ 59.94006 Hz = 79.98% blended
24/1.001 fps @ 60.000 Hz = 79.72% blended
24/1.001 fps @ 75.000 Hz = 63.74% blended
24/1.001 fps @ 120.000 Hz = 39.76% blended

Random partial v-sync offsets in the start of playback do not change the percentage of blended frames noticeably, except for the theoretically perfect refresh rates. 24/1.001 fps @ 60/1.001 Hz changes to match the nearly perfect 59.94006 Hz with 79.98% blended frames when adding a 0.3 v-sync offset at the start of playback.

[ashlar42]

I did some experimenting in Excel:
24/1.001 fps @ 59.900 Hz = 80.04% blended
24/1.001 fps @ 60/1.001 Hz = 40.12% blended
24/1.001 fps @ 59.94006 Hz = 79.98% blended
24/1.001 fps @ 60.000 Hz = 79.72% blended
24/1.001 fps @ 75.000 Hz = 63.74% blended
24/1.001 fps @ 120.000 Hz = 39.76% blended

Random partial v-sync offsets in the start of playback do not change the percentage of blended frames noticeably, except for the theoretically perfect refresh rates. 24/1.001 fps @ 60/1.001 Hz changes to match the nearly perfect 59.94006 Hz with 79.98% blended frames when adding a 0.3 v-sync offset at the start of playback.I really lost you here. :D

From the experiments in Excel I would conclude that it's better to playback 23.976 material at 60/1.001 but how can only 40% frames being blended? Are some frames simply repeated?

[iSeries]

24/1.001 fps @ 60/1.001 Hz = 40.12% blended
24/1.001 fps @ 59.94006 Hz = 79.98% blended


I don't really understand this (maybe I'm being dim), isn't 60/1.001hz = 59.94006hz?

I had my TV at 60hz, and I typed in Madvr allowable refresh rates of 60p and 59p, and used a 24/1.001 source with smooth motion. MadVR switched to 59p (59.94hz) to play it. So I assumed MadVR thinks 59.94 is better than 60 when using smooth motion.

[nevcairiel]

I really lost you here. :D

From the experiments in Excel I would conclude that it's better to playback 23.976 material at 60/1.001 but how can only 40% frames being blended? Are some frames simply repeated?

Don't think you can really get 40% only, not at 60 fps. Only every 5th frame has matching source/VSYNC timings, no matter how perfect the clock.

If you assume perfect timing, and 24 fps and 60Hz for simplicity (in milliseconds)

Frames: 0 41.66 83.33 125.00 166.66
VSYNC: 0 16.66 33.33 50.00 66.66 83.33 100.00 116.66 133.33 150.00 166.66 183.33 200


As you can see, a few frames fit perfectly on the timing, but only one in five on the output, so 4 in between will be blended (eg. 80%) - and thats already the best case.


I don't really understand this (maybe I'm being dim), isn't 60/1.001hz = 59.94006hz?

I had my TV at 60hz, and I typed in Madvr allowable refresh rates of 60p and 59p, and used a 24/1.001 source with smooth motion. MadVR switched to 59p (59.94hz) to play it. So I assumed MadVR thinks 59.94 is better than 60 when using smooth motion.

The key is that 60/1.001 is perfect precision, and 59.940 is not. As soon as it diverges just a tiny tiny bit (which you will), you might as well use 60Hz, since it no longer matters.
Unless SmoothMotion accounts for that of course and allows tiny deviation without blending. In general, just don't think too much about it. Using a native 24p mode is IMO always superior, but if your screen doesn't have that, SM is probably the best you can do.

[huhn]

@Asmodian

not sure how you used excel to come up with these numbers but madVR blends as little frames as possible. because no clock is perfect every vsync where two frames overlap is blended with other words it blends a frame every time a new frame needs to be displayed.

that means with 42 ms and 16 ms refresh rate 1 out of 3 v-syncs is never ever be blended and up to 2 out of 4 v syncs is not a blended frame so more than 66% blended frames is not possible and less than 50 % too.

with perfects clocks it would be 20% BTW.

[Asmodian]

I don't really understand this (maybe I'm being dim), isn't 60/1.001hz = 59.94006hz?

I had my TV at 60hz, and I typed in Madvr allowable refresh rates of 60p and 59p, and used a 24/1.001 source with smooth motion. MadVR switched to 59p (59.94hz) to play it. So I assumed MadVR thinks 59.94 is better than 60 when using smooth motion.

That was an example to show that while theoretically it is possible to get below the normal number of blended frames by tuning the refresh rate perfectly it isn't actually possible (59.94006 is an error of only 0.0000001%).

@Asmodian

not sure how you used excel to come up with these numbers but madVR blends as little frames as possible. because no clock is perfect every vsync where two frames overlap is blended with other words it blends a frame every time a new frame needs to be displayed.

that means with 42 ms and 16 ms refresh rate 1 out of 3 v-syncs is never ever be blended and up to 2 out of 4 v syncs is not a blended frame so more than 66% blended frames is not possible and less than 50 % too.

with perfects clocks it would be 20% BTW.

I see, I had some trouble trying to figure out an equation that gave reasonable results, I thought I got it but it still isn't accounting for the timing of the frame blends correctly so it was double counting blends. :o

I will see if I can figure it out... simple math is too easy! Must get simulator working! :p

Edit2: Hmm, frame 3 is displayed for 3.5 v-syncs? The blend ratios are still off. :mad:

Edit3: OK, I think I figured out something. All my logic is timed based now so I hope it matches what madVR calculates. I do get <50% blending but I believe this is correct? All frames total up to exactly the display time ratio we expect. I also calculated a new stat, "Average Blend" which is the average blending ratio across 20 thousand frames. I think this is actually a better representation of how much blending is going on, compared to percent blended frames.

24/1.001 fps video with smooth motion displayed at:

24.00000 Hz = 99.80% blended, 50% average blend
30.00000 Hz = 79.82% blended, 40% average blend
48.00000 Hz = 49.85% blended, 25% average blend
59.94006 Hz = 40.03% blended, 20% average blend
60.00000 Hz = 39.86% blended, 20% average blend
75.00000 Hz = 31.87% blended, 16% average blend
120.0000 Hz = 19.88% blended, 10% average blend
144.0000 Hz = 16.62% blended, 8% average blend

An interesting note about the average blend is that a V-sync offset doesn't change it at all unless you are using a nearly perfect refresh rate, and even then that is only because I am limited to 20,000 frames, as you can see the blending ratios drift up and down as the frame rate and V-sync drift in and out of sync.

I hope someone else finds it useful, it took me a while to figure out the spreadsheet. :o
http://i1222.photobucket.com/albums/dd496/asmodian3/Smooth%20Motion%20-%20Excel_zpstbob9nlx.png

[ryrynz]

Ya know, I always thought the higher the hz the more likely it was to have more blended frames.. Would if be wise to allow madvr the ability at higher refresh rates to change the ratio?

[Asmodian]

Change which ratio?

[ryrynz]

The amount of blended frames so its smoother. Couldn't you do approx half blended half repeated for example?

[huhn]

where do you get this info it is the % of blended frames goes down with higher refresh rates.

[Asmodian]

I just spent a long time attempting to quantify exactly how much less blending happens as the refresh rate increases. :p

24.00000 Hz = 99.80% blended, 50% average blend
30.00000 Hz = 79.82% blended, 40% average blend
48.00000 Hz = 49.85% blended, 25% average blend
59.94006 Hz = 40.03% blended, 20% average blend
60.00000 Hz = 39.86% blended, 20% average blend
75.00000 Hz = 31.87% blended, 16% average blend
120.0000 Hz = 19.88% blended, 10% average blend
144.0000 Hz = 16.62% blended, 8% average blend

My Excel Spreadsheet (http://www.mediafire.com/file/zvfvwtg3d545d9f/FrameRate_Mismatch.xlsx) if anyone is interested or wants to check my work. :)

[Q-the-STORM]

The amount of blended frames so its smoother. Couldn't you do approx half blended half repeated for example?
The point of madVRs smooth motion is to make it as smooth as 24Hz, not even more smooth than 24Hz. If you want it smoother, look into things like SVP...

[ddjmagic]

When using FSE mode with seek bar enabled, should I see a pointer once I move the mouse to the bottom of the screen bringing up the seek bar?
I see the seek bar but I don't know where the pointer is to actually click and seek, or is the seek bar just intended as a reference point to look at?

[burfadel]

Cursor should show, try updating your display driver. Latest are 17.7.2 for AMD and 384.94 for Nvidia.

[ryrynz]

The point of madVRs smooth motion is to make it as smooth as 24Hz, not even more smooth than 24Hz. If you want it smoother, look into things like SVP...

24Hz is not smooth though. The point of smooth motion is to make things more smooth and it was my opinion based on madshi's post about it that higher refresh rates lead to more blends which meant better motion resolution and smoothness.. but if Asmodian's calculations are correct then what madshi said about higher refresh rates being better is actually linked to the fact that's only because smooth motion is doing less.

Anyway, don't recommend SVP as an alternative to smooth motion, they're quite different and IMO not comparable with one another, also I dislike its results.

[vivan]

24Hz is not smooth though. The point of smooth motion is to make things more smooth"Smooth motion" emulates 24hz refresh rate on a screen with a different refresh rate. It does so by blending adjacent frames, which is basically pixel response time increased by 1/refresh rate (for some frames, not all).

Does watching video on a screen with 16ms response time makes it smoother? Kinda, but it's a very questionable thing to do.

[ddjmagic]

Cursor should show, try updating your display driver. Latest are 17.7.2 for AMD and 384.94 for Nvidia.

I've tried 384.94 (I was a version behind) but same thing. I actually don't get a mouse cursor at all in FSE.

[bcec]

Anybody having issues with 3D playback? When fullscreen, my 1080@24p frame-packed playback is playing as if contrast is cranked up 1000%. Only chroma upscaling is enabled.

EDIT: Turning off OS level HDR setting solved the issue. For some reason, Full Screen exclusive 3D (1080@24) + OS HDR turned on causes extreme contrast and clipped highlights and shadows with madVR. Maybe a bug?

[Q-the-STORM]

24Hz is not smooth though. The point of smooth motion is to make things more smooth and it was my opinion based on madshi's post about it that higher refresh rates lead to more blends which meant better motion resolution and smoothness.. but if Asmodian's calculations are correct then what madshi said about higher refresh rates being better is actually linked to the fact that's only because smooth motion is doing less.

Anyway, don't recommend SVP as an alternative to smooth motion, they're quite different and IMO not comparable with one another, also I dislike its results.

The point of smooth motion is to make 24fps video at 60Hz smoother than it is, but not smoother than 24Hz would be... the point of smooth motion is to get as close to what 24Hz looks like on any given Hz (assuming your video is 24fps)...
Yes, 24fps is not smooth and smooth motion is not there to make it smoother than the framerate of the video...

at 60Hz it might look/feel a bit smoother because the alogrithm can't perfectly mimic 24Hz (which is impossible), but if madshi could code something that would perfectly show 24Hz at 60Hz, that is exactly what smooth motion would do.... it's not there to make 24fps video smoother, it's there to remove the additional jerkiness that 24fps video at 60Hz (or other Hz mismatches) produces, to get as close to the "perfect" 24Hz as possible... that's why smooth motion doesn't do anything if you play 24fps video at 48Hz, 72Hz, 96Hz or 120Hz, because in those cases, you see every frame the exact same amount of time you would see them at 24Hz (assuming your display doesn't fuck something up)

That's why I brought up SVP, because it is doing what you are expecting from smooth motion, to make it smoother than the original framerate....

SVP and smooth motion have very different objectives and are not alternatives for each other....
what you are expecting from smooth motion is the exact goal of SVP - video that is smoother than the original framerate.... it's just that SVP doesn't do it by frame blending...
though if I remember correctly, you can set SVP to frame blend instead of using it's regular algorithm, which seems to be exactly what you are after, frame blending as much as possible, to get the smoothest video, without getting the artefacts that motion interpolation algorithms introduce...

[ryrynz]

My point is that anything other than frame blending is doing more than just making presentation smoother. I'm happier with the TV's interpolation rather than SVP.

[ryrynz]

Yes, 24fps is not smooth and smooth motion is not there to make it smoother than the framerate of the video...


And yet this is exactly what it does.. There is a night and day difference between me running at near perfect 24hz (1 frame drop per 8hrs etc) and running at 60hz with SM.

[pirlouy]

I doubt it. I did not see differences between 24Hz and 60Hz SM. I think 60Hz SM is still better for 60Hz panel (it's not sure they can do real 24 or 48 Hz).
For 120Hz panel TV, "near perfect 24Hz" + TV interpolation (not too much to avoid soap opera) is so perfect: no soap opera + no "24fps judder".

[ryrynz]

I can clearly see it on my TV regardless, it's obvious on pans.

[digitech]

And yet this is exactly what it does.. There is a night and day difference between me running at near perfect 24hz (1 frame drop per 8hrs etc) and running at 60hz with SM.

and which do you like the most?

[ryrynz]

and which do you like the most?Depends on the content. For normal films I prefer the extra movement resolution so I go with 24hz, for Anime it'll vary but generally I'd go 60hz with older animated stuff as I have this sharpened and thinned so the resolution loss is mitigated but I get the much smoother object movement.

[Dorohedoro]

Ok I'm really happy right now and I want to give my thanks to madshi for this amazing rendered.

Finally I was able to emulate the motion I see on the theaters and it was really easy at the end. I'm using a custom 23.974hz refresh rate but with my projector (x5000) I can still see a distracting flicker/judder on bright scenes if there is panning, on fast motion scenes or in some actor close up scenes, I've solved the problem using madvr smooth motion "only if there is judder" option. Now the flicker is gone, but also with this refresh rate there is no visible ghosting and the motion is flawless just like in my local theater, thank you so much and thank good I had enough patience to keep triying different combinations in the options because I hate 60hz and also the ugly ghosting smooth motion has in 60hz, and I really hate frame interpolation.... but if you try that option in 24p the outcome is amazing (remember to use "If there is judder" option under smooth motion settings).

Thank you so much madshi for giving us so many options to play with.

[nevcairiel]

I can clearly see it on my TV regardless, it's obvious on pans.

If the video appears smoother in 60Hz+SM then an actual 24Hz mode, then I would draw the conclusion that your TVs 24Hz mode isn't quite perfect (which unfortunately a bunch of TVs get wrong).

As others have said, SM isn't a replacement for the interpolation TVs perform, or SVP, its just a way to avoid motion judder from mis-matched refresh rates.

[70MM]

Finally I was able to emulate the motion I see on the theaters and it was really easy at the end. I'm using a custom 23.974hz refresh rate but with my projector (x5000) I can still see a distracting flicker/judder on bright scenes if there is panning, on fast motion scenes or in some actor close up scenes, I've solved the problem using madvr smooth motion "only if there is judder" option.

Im not sure if that is correct.
Im sure everyone here says that SM will never work on any of its settings if the framerate is set to 23/24Hz and everything in between.

Its easy to see, by checking in the onscreen log, it will state Smooth Motion is off, even though you have "only if there is judder ticked".

[Asmodian]

but if you try that option in 24p the outcome is amazing (remember to use "If there is judder" option under smooth motion settings).

If you use "only if there would be judder without it..." with 23.976 fps content and a 24Hz display smooth motion will simply be off while with a 60Hz display it would be on.

None of those options change the behavior of smooth motion, only when it turns on or off. On a 24Hz display and smooth motion set to "only if there would be judder without it..." the output is the same as disabling smooth motion entirely.

[QBhd]

Dorohedoro's post is an excellent example of the placebo effect. It's truly amazing what the mind can tell it's self :)

QB

[x7007]

When watching 4K HDR movie, does it suppose to be 8 bit or 10 bit ?


http://imgur.com/GWabH4f

I'm using Nvidia 382.53 drivers with HDR in windows disabled. When I enable it, it's become 10 bit

Everytime I enable HDR and run the movie my screen turn black, the sound keep playing but I can't see anything, changing the nvidia settings from the Default color settings to nvidia Color settings and using HDR in windows YCbCr 422 12 bit limited works fine. but if it's on default it causes black screen . is it a issue with the 382.53 ?

[Dorohedoro]

I dont know guys, but without smooth motion there is flicker everywhere, maybe is because my custom refresh rate I dont know, but I can assure you, it's working, for me anyways...

[ddjmagic]

I've tried 384.94 (I was a version behind) but same thing. I actually don't get a mouse cursor at all in FSE.

After messing with some settings in Windows 8.1, it seems that it's something to do with the scaling settings, my resolution is 3840x2160 and I have it scaling set to 250% (Default) if I drop it below 200% I can see the mouse cursor again in FSE anything over that and I lose it.

Is this a bug in 8.1? Can anyone suggest a workaround - it's hard to read anything if I drop the scaling down.

[mrcorbo]

When watching 4K HDR movie, does it suppose to be 8 bit or 10 bit ?


http://imgur.com/GWabH4f

I'm using Nvidia 382.53 drivers with HDR in windows disabled. When I enable it, it's become 10 bit

10-bit is correct.

Everytime I enable HDR and run the movie my screen turn black, the sound keep playing but I can't see anything, changing the nvidia settings from the Default color settings to nvidia Color settings and using HDR in windows YCbCr 422 12 bit limited works fine. but if it's on default it causes black screen . is it a issue with the 382.53 ?

Default is probably trying to send to your display a format that it doesn't support. Since it works when you specify the format, just do that.

[Dorohedoro]

Im a moron, I had the refresh rate set to 59.975 that's why the smooth motion is working....

[omw2h]

damn i love madVR

i have two specific questions,

my tv support 4:2:2 chroma subsampling,
due to the fact that half of the pixel averages the color value,
do i still benefit from higher chroma upscaling algorithms for the original video size and/or doubling the video size?

dithering..
if there are dither patterns the "reduce banding artefact" does its magic,
if i see noise it doesn't get better of worse with any dither option,
i don't see any difference while switching options or disabling dithering,
is there something i can do about it? (no noise reduction is set in my drivers or TV)

[Asmodian]

Im a moron, I had the refresh rate set to 59.975 that's why the smooth motion is working....

Ah good, that makes sense. I recommend you keep using those settings. The flicker goes away at 60Hz and smooth motion, as you have seen, keeps the motion quality the same as a in a theater. :)

Use control-J to toggle madVR's OSD to see what it is doing and trust your eyes when judging quality.

my tv support 4:2:2 chroma subsampling,
due to the fact that half of the pixel averages the color value,
do i still benefit from higher chroma upscaling algorithms for the original video size and/or doubling the video size?

dithering..
if there are dither patterns the "reduce banding artefact" does its magic,
if i see noise it doesn't get better of worse with any dither option,
i don't see any difference while switching options or disabling dithering,
is there something i can do about it? (no noise reduction is set in my drivers or TV)

Chroma upscaling is less important for you but it will still affect the output.

"reduce banding artifacts" happens before dithering, everything in madVR happens before it dithers. Any changes noticed due to it are not changing the dithering as done by madVR.

Dithering is pretty subtle, you hopefully wouldn't notice it with a display at 8 bit or higher. However, if your display or GPU is doing a conversion (such as to converting to 4:2:2 internally) it will mask any changes in madVR's dithering. Stick to ordered dithering if you don't notice a difference, it is fast and good quality. There is no reason to use error diffusion if you cannot see an improvement but disabling dithering is always a bad idea.

You can also temporarily set your display to 4-bit or even lower, to see what changes in the dither pattern but make sure the "trade quality for performance" option "don’t use linear light for dithering" is unchecked. This is mostly for curiosity as optimal at 4-bit might not be optimal at 8-bit.

[dbcooper]

Hmm, the new AMD Vega GPUs have new instructions for video processing:

http://i.imgur.com/VwY2dcPl.jpg (http://i.imgur.com/VwY2dcP.jpg)

[Goeggel]

I have been looking through the forum, but haven't found any solution yet: when using madVR, I always only get D3D9, not D3D11.

Zoom Player 13 with madVR 0.91.11
LAV 0.70.2
Windows 7 64bit
Nvidia GTX980 with driver 384.76

Windowed, I get: D3D9 old path
With FSE, I get: D3D9 new path

I think I have chosen all relevant configurations for D3D11, but I am not 100% sure. Anything I should test/check/change?

[huhn]

go to rendering -> general settings -> "use direct3D 11 for..."
you changed some other settings too but you should now that better than me.

[Goeggel]

go to rendering -> general settings -> "use direct3D 11 for..."
you changed some other settings too but you should now that better than me.

Many thanks for your prompt reply!
I had already enabled this setting. Somehow it doesn't work. Is there a debug log or any other way to check why D3D11 doesn't kick in?

[Asmodian]

Do any of madVR's settings work?

[Goeggel]

Yes, to me it looks like other settings work just fine. For example "switch to matching display mode" works on my TV. Or changing brightness under "color & gamma" works, too.

[huhn]

there is a debug log but it is not really readable for us.

and you change other settings too which may make trouble like present frames in advanced.

i recommend to use the default settings and try d3d11 again.

[MS-DOS]

Anything I should test/check/change?

Seems like you have desktop composition disabled. D3D11 presentation doesn't work without it.

[JayeDog]

i just updated to windows 10 creators update but can't see the "HDR and advanced color" option under display properties.

I have a gtx 950. Do I need a hdr compatible display for the option to show up?