madVR - high quality video renderer (GPU assisted)

[EpsilonX]

Yeah, that's normal.

First, bicubic 75 AR is used to upscale base chroma to the source resolution to produce 4:4:4. Next, when madvr calls for image to be doubled, NGU is used for luma, but CR ar is used for the chroma channel to upscale the 4:4:4 source to whatever output resolution you wanted.

For me, the only confusing thing is that NGU uses image upscaling settings to determine how to double the chroma, which is a bit counter-intuitive.

this is all correct...

chroma upscaling only upscales chroma to luma resolution... if there is additional scaling it's gonna use the image doubling and image upscaling options...

so what you got there is a 720p video on a 1080p screen...
luma is 1280x720 which is doubled to 2560x1440 with NGU Low and then downscaled to 1920x1080 with Bicubic 150 AR

chroma is 640x360 which is upscaled to 1280x720 with Bicubic 75 AR and then upscaled to 1920x1080 with Catmull AR

I think that the main reason of all these misunderstandings is the fact that when the chroma doubling is unchecked in the "image doubling" page, the chroma doubling is handled by an algorithm selected in the "image upscaling" page. Moreover, there is the "chroma upscaling" page which is unrelated in this case. This is totally confusing and not explained anywhere except the depths of this forum.



Hats off to all of you for that explanations... :thanks:
And yes, I never thought that Catmull will be used for chroma with my setting...

[Q-the-STORM]

Telion was on to something with changing the upscaling setting pages...

though I would mostly keep it like it is, make changes mostly about separating chroma out of image doubling and upscaling...
now you got chroma upscaling, image doubling, image upscaling and image downscaling...

right now chroma 4:2:0 -> 4:4:4 is determined by "chroma upscaling"
upscaling 4:4:4 chroma to display resolution afterwards is determined by "image doubling" and "image upscaling"...
upscaling luma to display resolution is also determined by "image doubling" and "image upscaling"...

Having 3 sections might be simpler:
image downscaling:
stays exactly the same, don't think anyone needs separate settings for luma and chroma downscaling...

chroma upscaling:
determines not only 4:2:0 -> 4:4:4 scaling but also additional scaling to display resolution... so if you set lanczos4 here, it will be used for 4:2:0 -> 4:4:4 scaling AND scaling to display resolution
the settings page itself would almost look completely the same, you just have to add double, quadruple (and octuple) options and make some special checkboxes for algorithms like reconstruction, since those can obviously only be used for 4:2:0 -> 4:4:4 upscaling...

luma upscaling:
the same as image upscaling was before, but now this only affects luma.... just doubling, quadrupling (and octupling) is added...


I think this way is a bit more straight forward, you got a clear separation between luma and chroma settings...

There is lots of space underneath the bars for sharpness, ringing etc, so the doubling settings might fit there even if you don't change the layout....
if you turn the radiobuttons for the upscaling alogrithms into a dropdown list, you'd get a lot of space back if needed...
though that would mean you got 2 consecutive dropdown lists, one for the algorithm and one for the strength of said algorithm...
same goes for "only if scaling factor" options, those could be turned into dropdown lists, but you'd also have 2 consecutive ones where one affects the other...
doesn't look as clean as mixing radiobuttons or checkboxes with dropdown lists..

[Asmodian]

The problem with seperate pages for luma and chroma upscaling is that there are many cases where the scaling is being done in RGB, you would need to overwrite the settings in a separate chroma upscaling page or gray out the entire page when not using image doubling. I think it is better to keep chroma and luma doubling options on the same page for this reason.

I do like the idea of dropping the separate activation settings for chroma, they probably do make sense in some situations but I bet most people don't use different ones for luma and chroma. Using a drop down for the scaling factor would be OK too, maybe use lots of dropdowns in image doubling to really pack in the number of settings on that page. ;)

How about a third drop down below the one for "quadruple chroma resolution" that is only active if chroma doubling is disabled with luma doubling enabled? This would make it obvious that it would only be used when scaling chroma to match the doubled luma (or the doubled+downscaled luma, depending on the resolutions and trade quality for performance options).

edit: Another option would be to not let you turn off chroma doubling if luma doubling was enabled but have really a lot of options in its dropdown, e.g. for Catmull-Rom, Lanczos, etc.

[Scyna]

So I just got this msi nvidia 1070 laptop and nnedi doesn't work for me. I just get a black screen and audio. I'm on 375.95 drivers. NGU runs fine for me.

[Georgel]

NGU still adds coil whine while no other algo does this with madVR.

[aufkrawall]

Not madshi's fault, nothing to be done about it.

[khanmein]

NGU still adds coil whine while no other algo does this with madVR.

chroma upscaling NGU-low & disable image doubling (no coil whine)

chroma upscaling NGU-low & enable double luma resolution NGU, low quality (coil whine)

chroma upscaling NGU-med/high & disable image doubling with any settings (coil whine)

chroma upscaling NGU-med/high & enable image doubling with any settings (coil whine)

regarding the coil whine is very soft pitch noise unlike those circumstances during gaming which hit above crazy fps.

[leeperry]

Yes. I'm seriously considering dumbing the whole upscaling/doubling logic down. I know some expert users will hate it. But looking through all the posts since v0.91.1 was released, almost every non-expert user got the settings wrong, even after I explained it multiple times.

Not only does the current logic allow for serious misconfiguration (which almost all non-expert users seem to fall for atm), but it also costs endless support and correction posts.

Maybe I'll find a way to make it all easier without losing too much functionality.
I've already thought about that. But leeperry would run amok if I did that... :p
I so much see it coming that soon SR+NGU won't be possible anymore but I've run more tests and I really like the combination of NGU mid for luma + NGU low for chroma + SR2 LL AR when it comes to motion smoothness in combination with 60Hz FRC, I'm looking forward watching that magic on the 144Hz BenQ EX3200R so pretty please don't break my bubble......if things have to go sour, could you please create a folder hack of some sort so I could still use NGU+SR together :scared:

I think the problem is that many ppl expect simple instructions for "best" PQ without doing their homework, but they overlook that many of those are both display and personal taste dependent, obviously an uncalibrated TN display(vast majority of computer screens) won't nearly need the same settings as a videoprojector for instance. We've all seen those "tutorials" that have it completely wrong(NNEDI3-128 for chroma yeah right), so yes maybe a setup wizard with simple/expert control panels would help. You are expecting the average newbie to have read hundreds of pages of this thread here on d9, they have not and they want the best right away as they can't sit and try even dang options combination.

[Uoppi]

I know madshi has mentioned this a million times already but just to be absolutely sure:

1. Only "reduce ringing artefacts" is recommended to be disabled with NGU but not "activate anti-ringing filter" (or "anti-bloating filter" for that matter)?
2. The above recommendation applies when using NGU for luma doubling but not when using NGU solely for chroma upscaling, correct? For example, when watching 1080p content on 1080p TV using NGU for chroma, "reduce ringing artefacts" might still be of benefit?
3. Should AR for upscaling and downscaling still be used in combination with NGU or should they be disabled too?

[HillieSan]

I choose the best scaler that is below the 10 ms. I am more sensitive to judder/jitter than to sharpness.

[Backflash]

[/LIST]And finally, Anti-Ringing is said to be included in NGU, but which of the ARs is/are the one/s to be disabled?
Reduce ringing artifacts can stay, I guess (different to anti-ringing)?
AR in image downscaling?
AR in image upscaling?
AR in upscaling refinement?
Any other AR?

Honestly, there are so many anti-ringing checkboxes, that I don't know, which are redundant, which are overkill or which are plain useless...

This is what i'm also confused about, with SR it's obvious when you disable it. But AR is very source dependent I'd have to check so many samples to know exactly which one works or not with NGU.

I'm not sure about weeding them out, but maybe condensing it in one category.
Or to simply grey out options that do not work/create too many artifacts with NGU(or any other algorithm) but not remove them for other algos.

[leeperry]

Indeed, I'm also unsure about the downscaler's AR and I think it harms PQ when SR2 LL's AR is engaged, need to run moar tests I guess.

There's also the story of whether feeding C-R with a sharper picture or Bicubic 100+ / SSIM 1D with a softer one, more tests to be conducted again.

[jerryleungwh]

As a non-expert user it took me quite awhile to finally be able to use the new algorithm properly and I want to apologize for asking the same question that so many people have asked before, should've read before commenting.

My experience so far is that the new algorithm is freaking awesome. I tried to push my GPU to the very limit and oh boy is it sharp. I use NGU-med for luma doubling to upscale a 720p anime to 1080p. When the animation moves too fast the sharpness is not as obvious but I think that is due to the artifacts and details lost during the downscaling process, but when the animation is still it is so sharp as if it is a 1080p video in the first place.

I also tried to use NGU-med for chroma upscaling and I think it looks quite a lot sharper than before when I used NNEDI32+SR1. I even turned off adaptive sharpen and I really don't see the image gets any less sharp than having it on. Thank you mashi for this incredible new algorithm!

[ryrynz]

I also tried to use NGU-med for chroma upscaling and I think it looks quite a lot sharper than before when I used NNEDI32+SR1

It is, it's too much for me even on low, we'll see how it ends up after it's been improved some.

[aufkrawall]

Why would it be too sharp? I think if we did a ground truth comparison for chroma, it should come closer to it than any other non-reconstruction based scaler.
But I agree that there may be other crucial factors in reality with chroma, aka distracting ringing in sources. But the same goes for luma, I personally won't blur everything just to mask the artifacts.

[jerryleungwh]

It is, it's too much for me even on low, we'll see how it ends up after it's been improved some.
Personally I think the extra sharpness looks really nice so I'm okay with it the way it is now

[Backflash]

Personally I think the extra sharpness looks really nice so I'm okay with it the way it is now

I also love it sharp, but i'm legally blind, so yeah, everything must be over sharpened hell for me to enjoy.

[Sunset1982]

Some more testing:

I exchanged my RX480 with my GTX1060 from my gaming pc.

With same settings and NGU high I can run 1080p to UHD with 34ms (GTX). With RX480 i got 48ms.

So I will stay with the GTX1060 until NGU got a better performance on my RX480.

[Ver Greeneyes]

By the way madshi, do you think you could add a way to toggle rather than just activate a specific profile, or alternatively add a hotkey to switch back to automatic detection? Maybe just turn 'activate' into 'toggle' so it switches the profile off if it's already been manually activated. I've set up a bunch of profiles for my laptop that I activate when I have to unplug it (so watching video doesn't drain power like crazy), but right now the only way to switch back is to reopen the video. Alternatively if you could add detection for when the device has AC power, that would be great too ;)

[HillieSan]

Some more testing:
With same settings and NGU high I can run 1080p to UHD with 34ms (GTX). With RX480 i got 48ms.


I think, 34 ms is still too slow.

[Uoppi]

Ah, I missed the earlier post(s) concerning AR confusion before posting my own.

Well, I'm sort of relieved I'm not the only one who's a bit confused (again). I'd take the "NGU for everything" route just to keep things simple (and due to serious lack of testing time) but different options may of course interact in so many ways. :scared:

[Gagorian]

I think, 34 ms is still too slow.

34 ms is not too slow for 1080p24 to UHD. As long as your queues don't go empty, it does not make a difference.

[HillieSan]

34 ms is not too slow for 1080p24 to UHD. As long as your queues don't go empty, it does not make a difference.

Interesting. I get dropped frames. I am not sure if this is related to empty queues. I upscale 1080p24 and 1080p60 to 4Kp60.

Yesterday I discovered something that is weird. I played a movie that showed 24 fps (source) and after changing some settings in madvr it changed to 59 fps. Why is this?

[Gagorian]

Interesting. I get dropped frames. I am not sure if this is related to empty queues. I upscale 1080p24 and 1080p60 to 4Kp60.

Yesterday I discovered something that is weird. I played a movie that showed 24 fps (source) and after changing some settings in madvr it changed to 59 fps. Why is this?

Are you using smooth motion or motion interpolation? Obviously if you play something at 60 fps that requires a lot lower average render time. Personally I dislike any motion interpolation and would just have the display swap refresh rate to match the source frame rate. If you're using HDMI it's also impossible to use UHD 4:4:4 10 bit at 60 fps, I believe it will fallback to 8-bit.

Increasing queue size will not help in your case, it might help if you rarely get a few dropped frames.

[HillieSan]

Are you using smooth motion or motion interpolation? Obviously if you play something at 60 fps that requires a lot lower average render time. Personally I dislike any motion interpolation and would just have the display swap refresh rate to match the source frame rate. If you're using HDMI it's also impossible to use UHD 4:4:4 10 bit at 60 fps, I believe it will fallback to 8-bit.

Increasing queue size will not help in your case, it might help if you rarely get a few dropped frames.

I use bluesky FRC to prevent the judder. I do not use smooth motion. I will experiment if a display swap to match the source frame rate will solve this problem.

I use 4:2:0 10 bits (or 12 bits)

[leeperry]

Turns out SR does add bloat(and SR's AB was removed for some reason I didn't understand IIRC) but SSIM 2D 100% LL with 50% AB tackles the issue very nicely, all roads lead to Rome yada yada ^^

I rest my case that SR LL's AR doesn't require downscaler's AR as the latter only worsens PQ, I ran all those tests on lowres SD(who knew 360x288@1080p could look so good!). I think 720p30 looks indeed better with vanilla NGU w/o SR, anyway my HD7850 doesn't quite have the horsepower to run it all together with SSIM on 720p so case closed. I'm reluctant to upgrade coz 280/290 boards provides poor efficiency, suck up a lot of power and 480's need a 64bit OS that I don't have and seem to provide poor NGU performance, anyway PQ is fantastic so the upgraditis monster can go back to where it came from :p

[HillieSan]

I use bluesky FRC to prevent the judder. I do not use smooth motion. I will experiment if a display swap to match the source frame rate will solve this problem.

I have enabled "switch to match display mode" and I disabled blysky frc. Now image doubling isn't working!! The render time goes up to 3 ms with NGU high and present goes to 25 ms or higher. Changing the luma doubling setting doesn't change anything. Am I missing something?

UPDATE: Of course, this keeps the display to 1080p and no image doubling is performed. My mistake.

UPDATE: Changing the frequency of the display from 60p to 24p doesn't help.

[Ver Greeneyes]

Smooth motion adds a lot of overhead for UltraHD output, thankfully you can detect it using a (smoothMotion) profile rule (it doesn't work without the parentheses) so I've set up a separate profile for it that doesn't use expensive scalers.

[MS-DOS]

RX 480 on Win 7 x64 with 16.10.3, 720x480 movie doubling (luma only), SXBR100 for chroma, No de-ringing and no SR.

NNEDI3-64: ~6.48ms ~8.1% GPU load
NGU High: ~8.45ms ~24.1% GPU load

Same GPU\Memory clocks in both cases. So, yeah, NGU is quite hungrier than NNEDI3.

madshi, I'm interested so see how it performs on D3D11 shaders.

[Gagorian]

I have enabled "switch to match display mode" and I disabled blysky frc. Now image doubling isn't working!! The render time goes up to 3 ms with NGU high and present goes to 25 ms or higher. Changing the luma doubling setting doesn't change anything. Am I missing something?

UPDATE: Of course, this keeps the display to 1080p and no image doubling is performed. My mistake.

UPDATE: Changing the frequency of the display from 60p to 24p doesn't help.

Set display to switch to 2160p23 (p23 for 23,976 hz).

[seiyafan]

For those of you upscale 1080 videos to 1440, have you noticed any benefit of using NGU (then with a downscaler) vs. Jinc?

[clsid]

Looking at the render times that various people have posted, maybe there is room for something in between NGU Medium and NGU High? With regard to naming, what about NGU low -> NGU1, NGU Med -> NGU2, etc?

[Asmodian]

For those of you upscale 1080 videos to 1440, have you noticed any benefit of using NGU (then with a downscaler) vs. Jinc?

I have, for animated content, the increased sharpness is pretty noticeable (bicubic 150 downscaling).

[har3inger]

Even NGU low is better for sharpness than jinc on 1080p->1440p, unless you crank SR for Jinc. High SR adds quite a bit of aliasing too.

@ Asmodian: I'm getting even better results with SSIM compared to bicubic 150. This could be entirely placebo on my part, however.

As for NGU on chroma, it is very sharp, which is fine (and desirable) on clean sources. If your sources have severe artifacts, which many rips will in the chroma channel, NGU will strongly emphasize them. Blocking becomes very obvious whenever it's present, whereas other scalers like reconsoft are able to hide most of it. Also, I've changed my mind about reconsoft vs bilateralsoftSR2; the former results in fewer artifacts and less flatness at the cost of slightly less sharp "pure chroma" elements like logos.

[madshi]

Is a performance improvement possible with DXVA2 native hardware decoder for NGU, instead of DXVA2 copy-back for AMD?
DXVA2 hardware decoding or not really doesn't have anything to do with NGU. Decoding and scaling are two totally separate steps. If anything, native DXVA2 handling could be improved, but only on Windows 8.1+ and it would require LAV to be extended quite a bit, too, so I'm not sure and (or even if) that will happen.

Quick question: can I save a "source white level" setting as default? It doesn't offer F2 to save as other options such as the "source levels" setting.
No, it's not possible. The "source white level" setting is only there to fix specific problems with specific videos, so I don't think it makes sense to be able to save this setting. The "source levels" is a special case because of which it is saveable.

why is NGU "next generation" ? and what benefits does it have over NNEDI3 ?
what can NGU do that NNEDI3 cant?
It reaches a whole new quality, IMHO, and especially a whole new quality vs performance ratio. I'll post a more detailed comparison with a speed vs performance graph, when work on NGU is complete, but that will probably take a couple of weeks.

its probably not as sharp as NNEDI3 but it still has more details i think.
it feels more that sharpness now is back to normal while NNEDI3 had a few % of artificial sharpness that isnt there from the source?
like comparing 8bit to 10bit
If you want to compare NGU to NNEDI3, turn off SuperRes for both. Then, with a reasonably sharp source, NGU produces much sharper results. If you enable SuperRes that changes the whole situation.

In terms of performance, I'm on a laptop with an OC'd R9 270x. NGU low runs substantially faster than SXBR 100 with SR2. Render time is 30 ms vs 40+ms. Image quality for the performance is astounding. It reminds me of when I first tried madvr and saw the difference between bilinear (ew, wtf) and lanczos image upscaling for the first time.
:)

The in-app performance meter says super-xbr is sharper than NGU. Is this true?
No, definitely not. The performance meter isn't updated for NGU yet.

What's consensus now on which options have to be disabled in conjunction with NGU?
I think it's too early for a consensus.

NGU already contains some amount of deringing, so you may not have to use the separate deringing algorithm in "artifact removal". Also, the separate deringing algo might make life harder for NGU, so image quality might even be better in many cases with it disabled. However, there's a chance that with some sources the separate filter might do a better job at removing ringing that NGU. So I can't give a final recommendation here.

Image enhancements and most upscaling refinements are not needed with NGU, unless you have a very soft source, maybe. The new "soften edges" and "add grain" options might be useful, if you like what they do.

Settings in chroma upscaling and image downscaling are not specific to NGU.

Hope I am not offtopic spamming but I found this. https://arxiv.org/pdf/1611.05250v1.pdf (Real-Time Video Super-Resolution with Spatio-Temporal Networks and Motion
Compensation) Very intereting and promising.

I was wondering if any upscaler in madvr is using future or past frames for more detail or if all real time scaler have to be spatial scalers only.
I've investigated temporal super-res years ago, and from what I've read it mostly only works if the source contains aliasing and isn't too soft. Looking at typical DVDs and Blu-Rays I think the sources we're going to playback are typically too soft to benefit much from a temporal super-res algorithm. To be honest, I'm not really a big expert on this topic, this is only what I gathered from some research a couple of years ago.

I think noise or compression artifact reduction has a better chance at making good use of past/future frames than upscaling.

As for NGU I am happy, altho I was too hyped after the initial teaser expecting quality near waifu2x. It still is good for low res anime, since its output is sharp and the cartoony effect is not disturbing with little aliasing.

Only problem I see is that many low res(anime) are low quality or old encodes from dvds with blocking or other artefacts, which can exaggerate with ngu.
To be fair, waifu2x has exactly the same problems with blocking or other artifacts as NGU has - except if you switch waifu2x into a noise reduction mode. Then waifu2x produces much better results than NGU. Maybe similar results/quality could be possible with NGU, too, when doing proper noise/artifact reduction before upscaling. I don't know, but something like this not likely to come to madVR soon.

2. The above recommendation applies when using NGU for luma doubling but not when using NGU solely for chroma upscaling, correct?
The "reduce ringing artifacts" option only works on the luma channel. So which algorithm you've selected in the chroma upscaling section doesn't matter.

3. Should AR for upscaling and downscaling still be used in combination with NGU or should they be disabled too?
AR for upscaling and downscaling is not a separate algorithm. E.g. there's "Catmull-Rom AR". But there's no "upscaling AR". Which means that this setting has nothing to do which image doubling algorithm you're using. IMHO the "AR" options in up- and downscaling sections should always be enabled for best quality.

madshi, do you think you could add a way to toggle rather than just activate a specific profile, or alternatively add a hotkey to switch back to automatic detection?
There is already a keyboard shortcut ability in the profile group settings which allows you to toggle profiles. Maybe you've missed it?

Turns out SR does add bloat
You can usually trust the madshi's eyes... ;)

(and SR's AB was removed for some reason I didn't understand IIRC)
It was removed because I couldn't see any benefit, and because nobody (including you) was able to provide any evidence to prove me wrong. SuperRes by itself usually *reduces* NNEDI3/super-xbr bloating. So the AB filter didn't really have anything (useful) to do.

It's quite possible that the SuperRes AB filter would now suddenly be useful, given that SuperRes now adds some bloating to NGU, which it never did before. But you know that I don't like activating SuperRes with NGU at all, so I'm not going to re-add SuperRes AB, if the only thing it would help is a situation that I don't recommend, anyway.

Looking at the render times that various people have posted, maybe there is room for something in between NGU Medium and NGU High?
Definitely. I will probably do that sooner or later. But I think there are things I should do first which would bring higher benefit to users.

E.g. what I'm looking at for next weekend is another quality step below NGU Low, which is again noticeably faster. I'm not sure it will work out ok, maybe quality will be too low to be useful, we'll see.

With regard to naming, what about NGU low -> NGU1, NGU Med -> NGU2, etc?
Might be an option. Or just "NGU" with a separate quality slider. I'm considering to change the whole upscaling/doubling settings page logic (to make things simpler), so I don't know yet what I'll end up doing.

[HillieSan]

Set display to switch to 2160p23 (p23 for 23,976 hz).

I did. Still judder.

[ashlar42]

Posting just to thank madshi for this new awesome development. I have done this in the past, but seeing his continued work and support is always something... beautiful. Beautiful is the right word.

Thank you.

[aufkrawall]

For those of you upscale 1080 videos to 1440, have you noticed any benefit of using NGU (then with a downscaler) vs. Jinc?
It's not worth the performance hit. If you just want Jinc to be a bit sharper, imho use SuperRes 2.

[JarrettH]

Speaking of NGU - my compliments:

- Edges blend in with one another without a change in brightness (eg. light colours on dark)
- Much less grain when compared with SuperRes, which tends to bring out coarseness in the picture
- Less of a "severe" look in details/texture, more effortless. IMO a natural look can only be achieved with less complication (no upscaling refinements)
- Finer lines

http://screenshotcomparison.com/comparison/191351

http://screenshotcomparison.com/comparison/191355

Even Family Guy at 480p looks more like HD when upscaled (I can run NGU med on SD and NGU low on HD)

I appreciate this development, madshi :thanks:

[leeperry]

It was removed because I couldn't see any benefit, and because nobody (including you) was able to provide any evidence to prove me wrong. SuperRes by itself usually *reduces* NNEDI3/super-xbr bloating. So the AB filter didn't really have anything (useful) to do.

It's quite possible that the SuperRes AB filter would now suddenly be useful, given that SuperRes now adds some bloating to NGU, which it never did before. But you know that I don't like activating SuperRes with NGU at all, so I'm not going to re-add SuperRes AB, if the only thing it would help is a situation that I don't recommend

Fair enough, I was only using SR2 LL at the time and I now realize how useful SR's AR is on ringing sources but anyway I don't like forcing all AR and AB sub-options everywhere, the same way I never liked those sxbr-AB variations for instance so SSIM 2D 100% with 50% AB seems to work beautifully :)

It takes a good while to find winning combos but once you've nailed them, it's a real blast.

I think it would be nice if you could add preexisting rules for SD/720p/1080p/1080p60 and so on, this would make it a lot more convenient than having to figure out stuff like if (srcHeight = 1080) and (Fps > 51) "1080p60" but I'll do my homework as I now need to create a whole bunch of extra rules.

I will also need to make a rule to always force FRC except for 60p, I wish there was a "if the refresh rate is identical to the movie frame rate" but fair enough, I'll figure out a rule too as I want FRC for 30p.

[kasper93]

It's not worth the performance hit. If you just want Jinc to be a bit sharper, imho use SuperRes 2.

I guess it is personal preference. I like the sharp look of NGU and I find it very good for 1080->1440 upscaling. Especially for high quality source where any sharpness lost is noticeable. Tho it depends on your hardware and eyes in the end.

[HillieSan]

I will also need to make a rule to always force FRC except for 60p, I wish there was a "if the refresh rate is identical to the movie frame rate" but fair enough, I'll figure out a rule too as I want FRC for 30p.

If I set the display to 23 or 24 fps for a 24 fps movie then I still get judder. Same problem with 59 and 60 fps. Only bluesky fps solves this. Any idea what the problem can be?

[CarlosCaco]

NGU is great! Thanks Madshi it's a great addition...
i m o latest madVR version but can t adjust soften edges and add grain in small steps... only 1 or 2 or 3... in the latest version this not suppose to happen or not?
Other question... the use sigmoidal light option apples to the doubling algorithms too? Or only to the Upscale algorithms... in case of yes... what did you recommend for NGU?


Enviado do meu iPhone usando Tapatalk

[Ver Greeneyes]

There is already a keyboard shortcut ability in the profile group settings which allows you to toggle profiles. Maybe you've missed it?Ah, I misunderstood how that works - I figured it would toggle through all the profiles one at a time, but it just switches between manual and automatic. Thanks, this works!

[aufkrawall]

I guess it is personal preference. I like the sharp look of NGU and I find it very good for 1080->1440 upscaling. Especially for high quality source where any sharpness lost is noticeable. Tho it depends on your hardware and eyes in the end.
I like it too, but imho there is hardly any difference to Jinc + SR, except that NGU is slightly more natural. Not to speak of the performance requirements, A GTX 1080 OC likely won't be enough for NGU low + SSIM 1D with 1080p 60fps. My 1070 can't even downscale it with bicubic then when doing also chroma NGU low scaling.

[Kotik]

Hello,

Either I am doing something totally wrong or things are not supposed to work the way I expect them to.

I ripped countless of times my Avatar 3D BluRay disc with latest MakeMKV v1.10.2 and yet the 3D subtitles are not working properly, for that matter they are not working properly even when I directly play the movie from my disc.

The 3D effects of the movie most of the time cut through the subtitles, and it doesn't matter which subtitle language I am using.

I am using latest madVR build + latest LAV, I tried with both xySubFilter and with MPC-HC internal subtitle engine.

I remember a couple of months ago madshi mentioned that 3D subtitles are supposed to work with xySubFilter.

FYI I got a passive 3D TV and madVR 3D setting set to auto, the movies themselves do work fine it is just the subtitles in both the bluray disc and the .mkv rip are not really working properly.

So am I the only one with this issue? am I doing something wrong? or 3D subtitles were never meant to work with madVR + LAV?

Thanx in advance.

[Ver Greeneyes]

I like it too, but imho there is hardly any difference to Jinc + SR, except that NGU is slightly more natural. Not to speak of the performance requirements, A GTX 1080 OC likely won't be enough for NGU low + SSIM 1D with 1080p 60fps. My 1070 can't even downscale it with bicubic then when doing also chroma NGU low scaling.I don't know about the SSIM 1D part, but as far as upscaling to UltraHD goes my laptop's GTX 1060 3G is currently using NGO low luma doubling for 1080p60 -> 2160p60 (chroma doubling turned off, using 3-tap lanczos instead). I bet a GTX 1080 could squeeze in the downscale too. I did have to turn off the anti-ringing pre-processing, but that seems to be recommended anyway with NGU.

Edit: Catmull or Catmull AR might be a better choice of upscaler, I dunno. Lanczos 3 + AR is a bit too much for that particular profile though.

[seiyafan]

For those of you who use NGU for 1080 -> 1440, which downscaler do you prefer?
And for those Jinc + SR2 users, do you also turn on "scale in sigmoidal light" for Jinc? Also under SR options, do you also enable both LL and AR?
Thx!

[Asmodian]

For those of you who use NGU for 1080 -> 1440, which downscaler do you prefer?
And for those Jinc + SR2 users, do you also turn on "scale in sigmoidal light" for Jinc? Also under SR options, do you also enable both LL and AR?
Thx!

I like bicubic 150 and ssim. I tried to notice a difference between SSIM 2D 100 and bicubic for downscaling to above the source resolution but I cannot so I use bicubic 150.

I do use sigmoidal light for Jinc upscaling but most of the time upscaling is never used (The Titan XP has enough GPU power for NGU low chroma doubling with NGU high luma doubling).

I don't use SR anymore but I used to use AR but not LL.

[aufkrawall]

For those of you who use NGU for 1080 -> 1440, which downscaler do you prefer?

The difference for downscaling 4k to 1440p between SSIM & bicubic 150 is usually totally neglectable and doesn't justify the higher GPU costs imho.
It's more important to use anti-ringing filter (relaxed is enough imho).


And for those Jinc + SR2 users, do you also turn on "scale in sigmoidal light" for Jinc? Also under SR options, do you also enable both LL and AR?
Thx!
Sigmoidal light has less aliasing but more ringing, however ringing isn't any problem with AR filter for 1080p -> 1440p.
Anti-ringing option of SuperRes is pure placebo and LL is not much of a difference, but leaving it unticked should be more correct in most cases.