madVR - high quality video renderer (GPU assisted)

[ROKUGAN]

Originally Posted by Backflash

no, for one it doesn't support it
how fast do you want it to be? 1070 already crashes most ridiculous madvr settings

No it doesn´t. I have a Zotac 1080 AMP! Extreme OC @ 2100 Mhz with a i7 6700K @ 4.6 Ghz and a 4K TV and it´s still not enough to max NGU without dropping frames, as leo91531's excellent test has already proven:

http://forum.doom9.org/showthread.ph...40#post1787840

If I use SVP on top of that (which I love, because it´s miles better than smooth motion for my personal taste) then performance is much worse and you can´t go above NGU medium settings without dropping frames.

[Mistery 73]

Originally Posted by ROKUGAN

No it doesn´t. I have a Zotac 1080 AMP! Extreme OC @ 2100 Mhz with a i7 6700K @ 4.6 Ghz and a 4K TV and it´s still not enough to max NGU without dropping frames, as leo91531's excellent test has already proven:

http://forum.doom9.org/showthread.ph...40#post1787840

If I use SVP on top of that (which I love, because it´s miles better than smooth motion for my personal taste) then performance is much worse and you can´t go above NGU medium settings without dropping frames.

What do you mean by SVP?

[ROKUGAN]

Originally Posted by Mistery 73

What do you mean by SVP?

SmoothVideo Project (SVP)

https://www.svp-team.com/wiki/Main_Page

To my taste, works much better than the interpolation settings in my TV (Auto Motion Plus = either insufficient or creates "soap opera" effect).
But it takes a toll in performance, specially paired with an also resource hungry madvr (which I love btw! )

[Mistery 73]

the smooth motion of madVR keep it off?

[huhn]

no it doesn't keep SVP off and SVP and smooth motion quite different smoothmotion can't create soap opera effect and SVP want to created soap opera effect(they stated this on there page...). SVP also want to make the source more smooth than playback at the correct refresh rate. smooth motion can't be more smooth than playback at the correct refresh rate.

for example using SVP to double 23 hz to 47 hz and using smooth motion to present this at 60 hz isn't a bad idea and it avoids the problems from a 5/2 interpolations.

[ROKUGAN]

Originally Posted by Mistery 73

the smooth motion of madVR keep it off?

You can keep it on, but it´s unnecessary with SVP so I keep it off. Follow this guide if you want to try it out:

http://www.overclock.net/t/1385468/s...cussion-thread

[ajp_anton]

Originally Posted by madshi

No, 35ms vs 42ms is not perfectly smooth at all. It might be smooth with a game but not with a video. It seems to me you don't really understand the difference. With a game, if you have 10ms -> 10ms -> 10ms -> 40ms, then a car moving from the left side of the screen to the right side, will move a little -> little -> little -> lot. Because of that it's not a big problem if the exact frame intervals vary a bit. With video the situation is different. The position of the car is already defined by the recorded video. If you have 10ms -> 10ms -> 10ms -> 40ms, that would be a catastrophe for video, because the car moves the same amount in each frame. For video we need *perfect* frame intervals. And unless G-SYNC can deliver that, it's useless for video.

Not advocating G-/Freesync, but I just want to say something here in its defense (or want to be corrected if I'm wrong). The game will pick a point in time to render its next frame, so in a sense the position of the car is "already defined" in the game as well as in the video. The difference is just that the game is flexible on what point in time it wants to render.
So a G-/Freesync game should be just as (un)smooth as G-/Freesynced video.

Also, things in the settings that still need to be clearer:
- "downscale quality" should also say what the algorithm actually is.
- "chroma quality", what is this and why is it active with NNEDI3 but not NGU?
- NNEDI3/NGU, "activate only if it's useful", what is being used to upscale when it's *not* useful?
- When scaling using DXVA, it should say that it also handles chroma (I'm assuming that it does).

[BluesFanUK]

I only tested this last night but I seemed to get better performance from NGU than NNEDI3 (On a 980Ti).

How is that possible over Pascal cards?

[zvans18]

http://screenshotcomparison.com/comparison/192201 (i think the glitches can be ignored, they only started when i disabled overlay to take the screens)

well i love sharpness, but even for me NGU is too destructive to even clean low res animation compared to nnedi3

i'd much rather put on a bit of edge thinning (not in screenshot) and put up with a little softness than deal with the mess NGU makes at the moment

[ispano]

@madshi v0.91.3

Feedback: Overall, a well done improvement cleaning house. Really appreciate that you've kept NNEDI3 even though I have shifted to NGU as my default for all profiles. I think the simplification was done in good taste even though I usually detest the limiting of choices. Maybe a "full/expert" versus "simple/minimal" is in order to avoid needless frametime troubleshooting and satisfy those who would like more control/tuning? The downscale pulldown menu in the "image upscaling" section is the only thing I find redundant and should probably be removed. Was the "image downscaling" section ever an issue? Keep up the good work madshi.

[Mistery 73]

on the advice of Rokugan I'm trying SPV4 but still do not know if for watching movies at 24fps is best use JRiver + mad and I think also use SPV4 put in crisis my video card you wsecondo worth using SPV4?

[Sp00kyFox]

Originally Posted by ajp_anton

So a G-/Freesync game should be just as (un)smooth as G-/Freesynced video.

with a game the agenda is to display a frame right at the moment it is completely rendered which is the desired behavior realized with g/free-sync.

with video though we already have finished frames and know beforehand when a frame is supposed to be shown. and according to madshi that's the whole problem. there is no API allowing you to tell when to display a frame. from what I understand with games the display sync with gsync is triggered in hardware by the completion of a frame rendering. you can't apply that to videos.

[StinDaWg]

Does anyone else have issues with Windows occasionally locking up to a black screen after exiting/closing a video while in fullscreen exclusive mode? This happened here or there in the past with D3D9 FSE, but with D3D11 FSE it seems to be happening more often (couple times a week). About 5% of the time after closing the player the screen will go black, and I will have to do a hard reset with the power button and reboot the pc.

As a test if I quickly cycle between windowed and FSE over and over occasionally it does the same thing and just crashes to full black screen.

Windows 10 x64
HD 7770 with latest AMD drivers
MPC-HC as player

[leeperry]

Originally Posted by StinDaWg

Does anyone else have issues with Windows occasionally locking up to a black screen after exiting/closing a video while in fullscreen exclusive mode?

I thought it was due to PotPlayer that I sometimes got a black screen for a second or two when closing it but creating a "madVR\DontRenderAfterStop" empty folder fixed it as far as I can tell. Prolly unrelated to your problem, though as I always was able to at least open the taskmanager and kill PotP.

[StinDaWg]

Originally Posted by leeperry

I thought it was due to PotPlayer that I sometimes got a black screen for a second or two when closing it but creating a "madVR\DontRenderAfterStop" empty folder fixed it as far as I can tell.

I will give that a try for now.

[ROKUGAN]

Originally Posted by Mistery 73

on the advice of Rokugan I'm trying SPV4 but still do not know if for watching movies at 24fps is best use JRiver + mad and I think also use SPV4 put in crisis my video card you wsecondo worth using SPV4?

See here, example:

https://www.youtube.com/watch?v=1MuFWCgDhgE

Wether you like the smoothness or not and if it's worth the extra GPU power needed (btw, I use the Intel integrated GPU on SVP and the Nvidia GTX 1080 for Madvr) is up to personal preference. In my case it's totally worth it, but many people doesn't like the smooth "videocam" effect and prefer the classic cinematic 24fps film cadence. I personally dislike the judder and find SVP to be more effective than the settings in my TV.

[marcobrd]

hi all, i'm a madvr user since many years and i really love it. i use it 90% for bluray upscale to 4k TV.
i like NGU very much and i can't think about using different algos from now on. i really want to thank you madshi for your incredible job.
i agree that the UI could be improved. IMHO there should be somewhere one "switch" option that allow users to switch to "basic" or "advanced" mode for the scaling section, maybe with a little hint of very few words which explains the difference in order to make it clear: the first one should be provided by default with presets for users who want to make it simple, the second one should be there in order to provide full control without limitations.

i noticed that d3d11 is about 2ms slower than d3d9 (win10 x64, geforce 1070, mpc-hc).

i have a question: with 91.1 if i set NGU on the same levels (say medium) for both chroma upscale and luma doubling, ctrl+j statistics only shows two NGU upscale processes (chroma, luma).
if i set NGU on different levels for chroma upscale and for luma doubling, ctrl+j statistics shows three NGU upscale processes (chroma, luma, chroma).
chroma doubling always disabled. 1080->2160 scaling.
could someone please explain why? is it better to have only one chroma upscale process in terms of optimal performance and PQ?
with 91.3 it looks like not possible to set the same NGU levels for chroma and luma.
thank you!

[cork_OS]

Originally Posted by zvans18

http://screenshotcomparison.com/comparison/192201 (i think the glitches can be ignored, they only started when i disabled overlay to take the screens)

well i love sharpness, but even for me NGU is too destructive to even clean low res animation compared to nnedi3

Good sample of NGU oversharp. At my point, NGU is more sharp/dirty than NNEDI3 [on not perfectly clean sources] even when soften edges enabled.

[pirlouy]

@zvans18: You did not put the original image, so your comparison has less value. There are known case where NGU has limitations, but if you don't post original image, madshi can't analyse where his algorithm can be improved.

For those who make comparison, don't forget the original image.

Edit: Good job at not doing it ispano. xD

[ispano]

Setup for comparisons of NGU

I know madshi has implemented a warning for SuperRes but also note this should apply to the "artifact removal" section for "reduce ringing artifacts". You will get aliasing with NGU and dehalo enabled and even stronger aliasing when dark dehalo is checked. It is also stated that dark dehalo is not recommend for anime. Just want to make sure NGU gets a fair shake when people are comparing screenshots. And yes, I agree NNEDI3 does a great job image doubling and striking a good balance. NGU already has Dehalo aka "reduce ringing artifacts" built in I believe which is causing the conflict. Aliasing inherent in the source material will of course be more obvious with NGU. That's also a good reason why NNEDI3 is a good fallback for sources where you want to hide those specific artifacts. Below is my 480p DVD source fed to Vapoursynth w/ AA processing.

NGU Very High - Waifu2x - NGU w/ dehalo - NGU w/ dark dehalo

NGU Very High

Waifu2x

NGU Very High w/ Deringing (DeHalo)

NGU Very High w/ Dark Dehalo checked.