madVR - high quality video renderer (GPU assisted)

[Nevilne]

Hmm, when playing back 30 fps on 60 it enables frc with first frc option, is that intended?

Originally Posted by Nevilne

There are some bugs with seeking.

Originally Posted by madshi

Could you be any more unspecific?

Sorry! If you seek during playback and if "delay playback start after seeking, too" is unchecked and "use a separate device for presentation" is checked, the previous frame will flicker for a few frames after seek.

Originally Posted by Nevilne

What would be suggested settings if I want to keep reclock just as audio renderer?

Originally Posted by madshi

Probably the first FRC option, I would say.

Hehe, I meant the reclock settings, i'm just worried that it might mess up playback speed somehow.

[cyberbeing]

In the next version could you add a setting to disable FRC for close matching refresh rates?

A) Disable FRC when refresh rate is within 0.001Hz of video frame rate multiple.

OR

B) Disable FRC when "dropped/repeat every..." calculation is greater than X hours?

OR

C) Disable FRC with video display frame rates greater than XXfps

OR

D) Disable FRC based on a complex refresh rate to frame rate exclusion list


My GT440 DDR5 (overclocked) needs to make too many quality compromises for 1440x1080->1920x1080 59.94fps deinterlacing, 1920x1080 59.94fps deinterlacing->1600x900 95.904Hz, and 1280x720->1920x1080 59.94fps @ 119.88Hz to be playable with FRC enabled. Horrible blurring artifacts occur whenever the render queue drops below half-full from my GPU being too slow. This is particularity a problem when entering fullscreen, since the queues empty and need to fill up again faster than realtime. Overall I don't seem to be seeing any smoothness benefit with FRC when using 23.976fps @ 95.904Hz & 23.976fps/29.97fps/59.94fps @ 119.88Hz, but as mentioned above the main problem is my GPU being too slow to do 1280x720/1440x1080 59.94fps/60fps -> 1920x1080 119.88Hz or 1920x1080 59.94fps/60fps -> 1600x900 with decent settings. In these specific cases my GPU was already being pushed very hard in previous madVR versions with minimalistic settings, so there isn't really any headroom for the extra GPU load from FRC.

Edit: The Keyboard shortcut for FRC may also be more useful as a Toggle rather than separate Enabled / Disable

[dansrfe]

This is my first time using the madTestPatternSource. I just looked at the SmoothMotion24.ytp file and madVR automatically switched the refresh rate to 48Hz, as I had specified in the refresh rate changer, but the lines at the bottom flicker quite a bit. Does this mean that the display is internally converting the 48Hz signal to 60Hz?

PS: I tested this with v0.85.8 (so no frp). Not sure if that makes a difference when using test patterns with madTestPatternSource though. Does it?

Also, what should I be looking for when viewing the colors.ytp and grayramp.ytp files?

Why is there an option in the smooth motion settings page for "... or if the display refresh rate is an exact multiple of the movie frame rate"? I thought that was a good thing... If 24 fps is being played back at 24Hz, 48Hz, 96Hz, or 120Hz then there shouldn't be any need for smooth motion frame rate conversion, right? I'm slightly confused.

Thanks madshi!

[trip_let]

Maybe this is the obvious and expected result, but I tested...

23.796 -> 59.94 playback
23.796 -> 59.94 playback with smooth motion FRC
23.796 -> 71.93 playback

on my nominal 60 Hz 1080p eIPS monitor. 71.93 is through custom timing in Intel control panel on HD 4000, which can't seem to do 71.928 exactly. FRC beats no FRC on 59.94 (blending looks pretty good overall, as in generally harder to notice and less offensive than juddering), but having an integer factor is better than that. I would say that in my limited impression so far, though there are different processes and things going on under the hood, subjectively I think the 59.94 FRC is closer to the 71.93 in "quality" than it is to the original raw 59.94.

Anyhow, very impressive work!

[Plutotype]

Originally Posted by madshi

I'm showing the original frames untouched as often as possible, but I'm showing blended frames as often as necessary to achieve the visual effect of a display with infinite refresh rate. There's no fixed pattern for specific framerate -> refresh rate combinations. The exact blending factors (and whether blending is needed at all) are calculated for every vsync interval. The blending factors usually change all the time.


No. Basically if you let madVR FRC 24fps content to 60Hz, the end result should look almost identical to displaying 24fps on a properly implemented 24Hz display.

So are there any motion sharpness or less flickering benefits when playing back 24fps on 60Hz ( or even higher refresh rates ) with FRC on?
Thanks for the new version anyway, I like this new feature.

[robpdotcom]

Small request: Can you make the keyboard shortcut for frc into a toggle? One shortcut to turn off/on, instead of one to turn on and another to turn off?

[Qotscha]

Originally Posted by Qotscha

I have similar problems when using dxva2 decoding. If I open a new file while playing a video in fullscreen exclusive mode on my secondary display, I often get a black screen or a MPC HC (or BE) crash. With software decoding everything works flawlessly.

0.86.0 fixed these problems

[n3w813]

I was skeptical at first when I read the release notes for 0.86.0 on smooth motion...I had to leave work early just so I can try it out at home.....

After testing, WOW, no judder at all with 23.976 sources running at 60Hz . As good as 24p with manually adjusted custom resolution settings plus Reclock. (I think it might be even smoother)

One downside though, I definitely see some loss of sharpness/clarity. Not much, but noticeable.

a bunch Madshi!!!

[pie1394]

Originally Posted by madshi

No, it's not a typo. Of course that only applies if FRC is turned on, which is not the default setting for 24fps -> 24Hz. But you can turn it on, if you want. The benefit would be endless playback with no dropped/repeated frames, without needing Reclock audio resampling. You'd lose sharpness, though. Basically whenever you turn FRC on, madVR simulates a display with infinite refresh rate by using frame blending.

Does the FRC option handle the Variable Frame Rate contents (Ex: 24p + deinterlaced 60p OP/ED anime) like this?

23.976 / 24 fps --> 23.976 / 24Hz (direct, no blending)
59.94 / 60 fps --> 23.976 / 24 Hz (motion blur)


madVR causes heavy playback stutters at the specified refresh rate when it presents the non-24fps contents in the middle of video files.

i.e. Japanese ISDB-T 24fps animes contents with horizontal scrolling texts. It looks like the broadcaster's encoding equipments are always to merge the scrolling text's video at 60-field rate instead of 24/30 full-frame rate.

[JarrettH]

Yes, there is an ever so slight reduction in sharpness. Maybe disabling anti-ring could "counter" that.

[cyberbeing]

Originally Posted by cyberbeing

Experienced a random freeze while seeking a WMV:

madVR 0.85.8
LAV Video 0.55.3 (Microsoft WMV DMO)
LAV Audio 0.55.3
WM ASF Reader

Freeze Report

Edit 2/19: Another freeze while seeking a WMV, but this time it hung MPC-HC entirely so a Freeze Report was unable to be generated. I'd suspect this is probably a LAV Video bug rather than a madVR bug, but the randomness of the freeze make it difficult to confirm.

Here is another freeze report seeking a different WMV multiple times in succession while the video was paused. Hard to say if it froze from the seek, or if it froze when I attempted to unpause right after the seek.

madVR 0.86.0
LAV Video 0.55.3 (Microsoft WMV DMO)
LAV Audio 0.55.3
WM ASF Reader

Freeze Report 2

So far I've been unable to reproduce this when using the Microsoft WMV DMO directly outside of LAV Video. Seeking speed also seems an order of magnitude faster using the WMV DMO directly. I'll continue testing over the next few days without LAV Video.

[agustin9]

the smooth option looks great!

[leeperry]

Originally Posted by madshi

Scaling is done *before* frame blending. So the cost of FRC is totally independent of the scaling algorithm. Meaning if you let madVR FRC a Blu-Ray movie to 120Hz, the cost of using Jinc3 AR is not higher than when using 24Hz. Of course there will be added cost for the frame blending and 120Hz presentation, but that's much lower compared to Jinc3 AR.

OK, thank you for the details!

Silly question here: frame interpolation(à la SVP) means that additional frames will be created in order to make the movie look smoother.....what you do is blend frames, how is that any different exactly

But well, I guess it might have to do with the "linear blending" that can be used for deinterlacing.....and my knowledge in this area is very poor

Also, while we're at it, ages ago ppl on HCFR were going crazy with the custom gamma curves that can be set in the nvidia drivers(and RivaTuner for AMD graphic cards), so with a lot of patience you could get the R/G/B gamma curves of your display to do whatever you want(they can set independently for R/G/B):



Too bad it's all done in 8bit, but you said that later on you might consider implementing your own CMS so I thought that this might a good feature request for a full blown 16bit CMS =)

Maybe you could just import those data from the nvidia drivers/Rivatuner registry, no need to reinvent the wheel really and they could be finetuned without running test patterns inside mVR(which is utterly time consuming).

[cca]

Anyone tested with DVD playback? I had trouble with the menus, with SmoothMotion ON static DVD menus do not update/display properly anymore. Animated menus work with a some additional lag.

[chainik_svp]

Originally Posted by mark0077

The only "problem" with this is its very cpu / gpu intensive, especially when the target rate is not a nice multiple of the display rate, so for me going from 23.976... to 59.996 is much more intensive than 23.976... * (5 / 2) = 59.94

I'm wondering do you think its a good option for those of us that want the smoothness of something like SVP with these nice multiples (to reduce cpu / gpu usage, or to increase SVP's quality settings), but with the accuracy of having something that helps us get 0 drops / repeats, could use SVP with your smoothing set to "always" to achieve the best of both? Is madVR theoretically going to kick in at those times when ordinarily this exact multiple 59.94 output from SVP, might not be accurate enough, and cause a frame repeat / drop on my display at 60.001 (which needs 59.996 to not drop / repeat anything).

I believe SVP with 59.94 + madVR's FRC will give you the same result as turning on "Blend frames to screen refresh rate" in SVP itself, the only difference is blending is done on GPU now.

[Anima123]

Originally Posted by madshi

Settings are stored in the registry, too. Please check "HKEY_CURRENT_USER\Software\madshi\madVR\Settings". MPC-HC should have write access to that, so madVR should be able to store its settings there. Not sure why that doesn't work for you. It seems to work just fine here.


Let us know what you find!

I cannot save madVR settings either after upgrading to the latest version with restricted priviledge of windows account. After delete the whole madVR registry, it also failed to recreate a new entry.

It's definitely a bug for me.h

[nevcairiel]

Originally Posted by cyberbeing

So far I've been unable to reproduce this when using the Microsoft WMV DMO directly outside of LAV Video. Seeking speed also seems an order of magnitude faster using the WMV DMO directly. I'll continue testing over the next few days without LAV Video.

The problem is with the WM ASF Reader, it sends a new media type on every seek, which causes LAV to re-init the decoder, which is just slower. However, as a consequence LAV is about the only decoder which can properly handle dynamic format changes.

I can probably optimize this a bit, try to detect if the format itself doesn't change at all and in that case just flush the decoder instead of re-creating it, but it'll be a while until i find time to look into it.

Stop using the WM ASF Reader?

[cyberbeing]

Originally Posted by nevcairiel

The problem is with the WM ASF Reader, it sends a new media type on every seek, which causes LAV to re-init the decoder, which is just slower.

Just so I'm clear, do you believe that WM ASF Reader's behavior was causing LAV Video to hang, and this isn't a madVR issue?

In that case, I'll try using LAV Splitter + LAV Video for WMV from now on.

Originally Posted by nevcairiel

Stop using the WM ASF Reader?

For some reason, the other day I looked for "microsoft" & "wmv" in LAV Splitter, but didn't make the connection that the format "asf" was used for WMV as well. Though in hindsight, I guess it should have been obvious considering the name of Microsoft's source filter.

[nevcairiel]

Originally Posted by cyberbeing

Just so I'm clear, do you believe that WM ASF Reader's behavior was causing LAV Video to hang, and this isn't a madVR issue?

I can't be entirely sure what exactly is causing it, but at least the WM ASF Reader seems to be a factor. I may have time to investigate more on the weekend, all i have to go from now are your freeze reports.

[zn]

Anyone tried to run madVR in Linux using Windows virtual machine with VT-d enabled direct access to PCI-Express GPU ?
( Xen VGA Passthrough / ESXi VMDirectPath / KVM / QEMU / spice)