madVR - high quality video renderer (GPU assisted)

[AngelGraves13]

Doubling the chroma channel is pointless. The Luma channel should be doubled and then all scaling should be done and then finally the chroma channel should be scaled. This will avoid needlessly doubling the chroma and then scaling it back down.

And no there is currently no way to do this I have tried multiple times. any form of doubling results and doubling of the Luma and chroma channels, even when scale chroma separately if it saves performance is checked.

what we need is a setting such as "scale chroma last."

[huhn]

change chroma to "normal" and madVR will use bicubic 60 AR.

of cause super XBr is always doing both at the same time.

[oddball]

Can anyone tell me how to use those HDR test patterns to work out my nits value? Also I would prefer to run my OLED55B6V in wide color gamut (Reds look more like red rather than orange red) with 'deep color' set on the HDMI input. However they are a little oversaturated on video. I know without calibration this is not ideal but is running DCI-P3 correct? I'm outputting 4:2:0 video to a 4:4:4 yCbCr pixel format (that's what the RX 480 chooses by default but full RGB seems better to my eyes when using DCI-P3 and way oversaturated in 709) and MadVR plays it at 10bit 4:2:0 FSE. My issue is it's either oversaturated reds and greens in 709 or some odd 'off' looking reds in DCI-P3 colorspace. BT2020 just looks straight out washed out. The whole thing is kinda confusing.

[Asmodian]

madVR always renders to RGB, having your GPU convert to 4:2:0, or even 4:4:4, after madVR is not optimal.

To get correct colors from madVR you need to tell it what gamut your display has. For the HDR Standard mode (don't use the other HDR modes) set "this display is already calibrated" to DCI-P3 on the calibration page, the OLED55B6V has 95% coverage of DCI-P3 in HDR. Sadly its greens are not quite right but that is the best you can do for wide gamut without creating a 3DLUT for it. :(

It is much more accurate in its SDR modes. For BT.709 content use ISF Expert (Bright Room) or ISF Expert (Dark Room) and tell madVR it is already calibrated to BT.709. Based on reviews this is how I would run that display, if I didn't have calibration hardware.

[StinDaWg]

Jesus. Might as well say that the real world impact of ANY of madVR's settings are close to nothing/placebo if that's the case.. you can quite clearly see the changes here even without enlargement. But sure, let's just remove all the chroma options.. it's not important, why have choices for this.. it's barely worth discussing. Madshi I vote to remove everything but reconstruction soft! :rolleyes:

Also I don't even know why you bothered to post :confused:
Maybe your memory isn't too good, most of us have been here since 09/10, I think we know what we're doing.
I'm sorry, but cutting out and focusing on small red text doesn't have much affect on real world viewing, that's all I'm saying. It's fine for a reference, but chroma is way overrated IMO. You will not convince me to waste gpu and use anything other than Bicubic 60AR, since I see zero difference in real world viewing. I never said any options should be removed however, just that it's not that important compared to other areas.

On the other hand, the differences with NGU doubling are immediately noticeable. No need for a magnifying glass and it's applicable to all sources.

I've been around using madVR and reading doom9 just as long as you (or longer) and am entitled to my opinion just like everyone else is. Irrelevant point.

[ryrynz]

I'm sorry, but cutting out and focusing on small red text doesn't have much affect on real world viewing, that's all I'm saying.


Did you check the original content? Seems not, because it's not exactly as small as you make it out to be. On a screen at optimal viewing distance this sort of thing is actually quite apparent and one of the better examples I've seen of how chroma affects the image.

Regardless of size in this example, these same effects are at work and give a clear guide as to how each algorithm works in this particular situation. Something other people interested in such differences who have requested such are interested in viewing. If you're not interested then just don't comment.


It's fine for a reference, but chroma is way overrated IMO. You will not convince me to waste gpu and use anything other than Bicubic 60AR, since I see zero difference in real world viewing. I never said any options should be removed however, just that it's not that important compared to other areas.


I personally don't think anyone really overrates chroma at all, almost everyone here is quite well educated over these 2000 odd pages with what to expect.

Does madshi overrate chroma by allowing NNEDI 256 neurons as an option, or future improvements for it with NGU?

Does Shiandow overrate it with his time spent on his bilateral upscaler?

Does feisty2 overrate it with his time spent on Chroma reconstructor?

These are options available to us from people who have invested considerable time into something they feel is important. It's wise to be knowledgeable about such options if they're provided regardless how how little they may affect the output in many situations.

These people spent the time to create these options and madshi has provided these to us after deeming them suitable as such, even hardware manufacturers like Panasonic specifically advertise their chroma upscalers even on newer UHD devices. To each their own. They have value in attaining high image quality and the proof is right there in front of you as I just highlighted.


On the other hand, the differences with NGU doubling are immediately noticeable. No need for a magnifying glass and it's applicable to all sources.


This isn't a chroma vs luma war. You're preaching to the choir here.


I've been around using madVR and reading doom9 just as long as you (or longer) and am entitled to my opinion just like everyone else is. Irrelevant point.


TBH I feel few people here share your view on the matter, also stating experience and knowledge is not irrelevant at all, it's quite pertinent to your comments actually.

[oddball]

madVR always renders to RGB, having your GPU convert to 4:2:0, or even 4:4:4, after madVR is not optimal.

To get correct colors from madVR you need to tell it what gamut your display has. For the HDR Standard mode (don't use the other HDR modes) set "this display is already calibrated" to DCI-P3 on the calibration page, the OLED55B6V has 95% coverage of DCI-P3 in HDR. Sadly its greens are not quite right but that is the best you can do for wide gamut without creating a 3DLUT for it. :(

It is much more accurate in its SDR modes. For BT.709 content use ISF Expert (Bright Room) or ISF Expert (Dark Room) and tell madVR it is already calibrated to BT.709. Based on reviews this is how I would run that display, if I didn't have calibration hardware.

In 4:4:4 full RGB I found that blacks were being crushed severely so had to go back to 4:4:4 yCbCr. DCI-P3 for 4:2:0 sources also caused problems with colors, so I set it back to BT.709. Everything looks right except the brightness of bright reds and greens being a bit glary causing fizzing and a bit of detail loss (that's the only way I can explain it). However all the bright primary colors looks as they should in wide gamut mode. Whenever I look at a display not in wide gamut mode reds in particular do not look red. They look kinda like a orange ruddy color. I also have a wide gamut HP display I use for my main PC display and since then I cannot go back to the old normal gamut. I just need to figure how to tone down the saturation a little. I have an old i1 Display puck and am wondering if the old (and free) HFCR with AVCD disc calibration method would work on OLED. I can use X-Rite on PC output and do it automatically, but it would then only set it for that HDMI input via the Windows colorspace config, which kinda defeats calibration on a TV (it really needs a calibration for all sources).

Most people don't notice such things but both my HP monitor and my previous Panasonic plasma I calibrated using the old method and kinda got used to it :)

[Sunset1982]

In 4:4:4 full RGB I found that blacks were being crushed severely so had to go back to 4:4:4 yCbCr. DCI-P3 for 4:2:0 sources also caused problems with colors, so I set it back to BT.709. Everything looks right except the brightness of bright reds and greens being a bit glary causing fizzing and a bit of detail loss (that's the only way I can explain it). However all the bright primary colors looks as they should in wide gamut mode. Whenever I look at a display not in wide gamut mode reds in particular do not look red. They look kinda like a orange ruddy color. I also have a wide gamut HP display I use for my main PC display and since then I cannot go back to the old normal gamut. I just need to figure how to tone down the saturation a little. I have an old i1 Display puck and am wondering if the old (and free) HFCR with AVCD disc calibration method would work on OLED. I can use X-Rite on PC output and do it automatically, but it would then only set it for that HDMI input via the Windows colorspace config, which kinda defeats calibration on a TV (it really needs a calibration for all sources).

Most people don't notice such things but both my HP monitor and my previous Panasonic plasma I calibrated using the old method and kinda got used to it :)

Oddball, if you need help calibrating an oled write me a pm. I could help you with that...

[Uoppi]

NGU low vs NGU very high
(http://screenshotcomparison.com/comparison/193089)


I find NGU low much nicer (less "dirty" and more even/smooth) than NGU very high in this example.

Serious, possibly dumb, question: am I "looking at the wrong things" or is it just a matter of preference (I guess it would mean I prefer softer scaling, right)?

[huhn]

i personally can see what you mean but in the end i'm not a fan of this "sample" anyway.

[burfadel]

For those with AMD and having issues, the 16.12.1 drivers are very much better with madVR. Everything works like it should!

[Uoppi]

i personally can see what you mean but in the end i'm not a fan of this "sample" anyway.

Yeah, was kind of surprised myself to see I clearly preferred the "inferior" option.

I suppose the very high setting is attempting to preserve more detail, hence also emphasizing possible artifacts. By my logic it would imply NGU very high would work best for high quality sources and NGU low for SD upscales. Though from what I understand, most (?) here think "the more passes the better" for NGU - and I was probably expecting the same.

[huhn]

For those with AMD and having issues, the 16.12.1 drivers are very much better with madVR. Everything works like it should!

d3d9 present in advanced is still broken. and NGU performance is still unacceptable slow on polaris.

power management is a lot better as least for me that's it.

i understand something else under "very much better"...

[burfadel]

d3d9 present in advanced is still broken. and NGU performance is still unacceptable slow on polaris.

power management is a lot better as least for me that's it.

i understand something else under "very much better"...

Ah ok, I just tried D3D11 Exclusive, seems to be working great! No doubt there are still issues, but at least I'm happy I can use madVR to its full potential now.

[ryrynz]

Serious, possibly dumb, question: am I "looking at the wrong things" or is it just a matter of preference (I guess it would mean I prefer softer scaling, right)?

It's hard to know how it should look exactly. I'm thinking much like how NGU very high displays it? Definitely not a dumb question. At the end of the day it's just one image that shows clearly visible changes between the algorithms, not obviously displaying what's more accurate or what affects the picture beneficially the most, which may be why huhn doesn't like it. I'll continue to look for better examples.

[HillieSan]

For those with AMD and having issues, the 16.12.1 drivers are very much better with madVR. Everything works like it should!

Ah ok, I just tried D3D11 Exclusive, seems to be working great! No doubt there are still issues, but at least I'm happy I can use madVR to its full potential now.

No improvements here with these new drivers. The HDMI diagnostics makes things buggy here.

Full potential? Have you enabled the dxva? dxva should be off.

[leeperry]

It's hard to know how it should look exactly.

Yeah, it's all moot without a 4:4:4 groundtruth.

[Paul Tronc]

The real world impact to any of those chroma comparisons is so close to nothing/placebo it's barely even worth discussing.

As madshi has said a million times, luma is way more noticeable/important.


Hello! This interesting. Being a noob, may I ask which settings are adequate for my needs, I mean optimize luma?

Currently I use NGU in this menu :

(sample image to illustrate the menu)
http://i1357.photobucket.com/albums/q760/Warner306/madVR%20Guide/Chroma-Upscaling_zpsya9y0gba.png

Should I set this off and use this feature : NNEDI image doubling ?

(sample image to illustrate the menu)
https://wiki.mikejung.biz/images/0/01/MadVR_image_double_nnedi3_gtx_970.png

I use madvr on 1080p sources on a 1080p projector on a huge screen.

[Warner306]

Hello! This interesting. Being a noob, may I ask which settings are adequate for my needs, I mean optimize luma?

Currently I use NGU in this menu :

(sample image to illustrate the menu)
http://i1357.photobucket.com/albums/q760/Warner306/madVR%20Guide/Chroma-Upscaling_zpsya9y0gba.png

Should I set this off and use this feature : NNEDI image doubling ?

(sample image to illustrate the menu)
https://wiki.mikejung.biz/images/0/01/MadVR_image_double_nnedi3_gtx_970.png

I use madvr on 1080p sources on a 1080p projector on a huge screen.

If 1080p -> 1080p, maximize chroma upscaling and maybe add debanding. That's really all you can do with a source like that.

[Neo-XP]

@madshi For image doublers :

Shouldn't Lanczos3 AR use the relaxed AR by default for luma downscaling ? I also thought it was your recommended setting.
I think you can remove all the LL variants from the luma downscaling after doubling drop-down list, LL is not good to use after doubling.

Can you instead add more choices for the "algorithm quality" for chroma ?
I'd like to use the same algorithm as the luma upscaling one after doubling, like it was in previous versions. Would that be possible ?

This and an option to disable the forced NGU quadrupling (meanwhile you work on a direct triple/quadruple upscaling), and I would definitely migrate to the next releases for good.

If this is not compatible with the new UI layout, can you define "Bicubic60 AR" as "low" and "Lanczos3 AR" as "normal" for chroma quality ?
I've found "Lanczos3 AR" to be the closest to NGU, and I don't want to waste resources on NGU for this chroma operation (and "Bicubic60 AR" is too soft).

[pirlouy]

Your impression is not enough. From what I see, Madshi reads scientific reviews to learn about new algorithms and progress in image improvment. When he makes a decision, there's a big chance it is the good one.
Don't forget the best image is not what you like the most, but the one which looks like the original the most. So we'll have to do ala Hydrogenaudio dictatorship: you need to provide proofs from your analyse with screenshots like this post (http://forum.doom9.org/showthread.php?p=1788868#post1788868) for example. Madshi said he won't add option for one guy, so you have to convince him !

[Shiandow]

Your impression is not enough. From what I see, Madshi reads scientific reviews to learn about new algorithms and progress in image improvment. When he makes a decision, there's a big chance it is the good one.
Don't forget the best image is not what you like the most, but the one which looks like the original the most.

While I agree that there are some 'misguided' ways to process an image, and that by knowing more about the theory behind the methods you can make some better judgements on what kind of processing is and isn't a good idea, let's not get confused about the end goal. The goal is to get the best image, and this is purely subjective (this isn't up for discussion by the way, it says so in the forum rules ;)).

All these heuristics like downscaling an image and upscaling it again with different algorithms to see which is closest to the original, are just that, heuristics. They are useful for comparing algorithms, or even designing new ones, but they're no substitute for actually looking yourself.

[AngelGraves13]

madshi, please add an option to DISABLE chroma doubling for those who don't want/need it in the drop-down list. I can run VeryHigh NGU, but the chroma doubling seems useless, even with the Normal (Bicubic 60) option. I'd just like to disable the chroma doubling altogether.

I'm going back to Jinc SR4 for Chroma and Upscaling. Looks just as good and has very low render times (15ms)

[jkauff]

While I agree that there are some 'misguided' ways to process an image, and that by knowing more about the theory behind the methods you can make some better judgements on what kind of processing is and isn't a good idea, let's not get confused about the end goal. The goal is to get the best image, and this is purely subjective (this isn't up for discussion by the way, it says so in the forum rules ;)).

All these heuristics like downscaling an image and upscaling it again with different algorithms to see which is closest to the original, are just that, heuristics. They are useful for comparing algorithms, or even designing new ones, but they're no substitute for actually looking yourself.
Gospel from Shiandow. Should be a sticky.

[AngelGraves13]

Really need some animated or real world content shot in 4:4:4..

It's easier to pick a winner when there's just one color like in above scenario but it may just be that more times than not NGU very high is the most accurate.
Some extra tests between the top contenders.

Super-xbr 100 AR vs NGU low (http://screenshotcomparison.com/comparison/193084)
NGU Low vs NGU med (http://screenshotcomparison.com/comparison/193085)
NGU med vs NGU high (http://screenshotcomparison.com/comparison/193086)
NGU high vs NGU very high (http://screenshotcomparison.com/comparison/193087)
NGU low vs NGU very high
(http://screenshotcomparison.com/comparison/193089)


NNEDI3 32 is quite nice and IMO looks better than NGU low but it's considerably more demanding at least on this HD4000, it's more demanding than NGU very high which makes it hard to recommend.
One wonders if perhaps it's time for NNEDI3 to retire from chroma duties.. That asks for one more I guess..

NNEDI3 32 neurons vs NGU very high (http://screenshotcomparison.com/comparison/193091)

How about Jinc SR4 vs NGU Very High for Chroma?

[khanmein]

d3d9 present in advanced is still broken. and NGU performance is still unacceptable slow on polaris.

power management is a lot better as least for me that's it.

i understand something else under "very much better"...

a lot AMD users said don't have issue & try ask them thru;

https://forum.lowyat.net/topic/4100453

http://www.overclock.net/f/67/amd-ati

this is y i'm NV fan boy!

[burfadel]

a lot AMD users said don't have issue & try ask them thru;

https://forum.lowyat.net/topic/4100453

http://www.overclock.net/f/67/amd-ati

this is y i'm NV fan boy!

Recent NVidia cards have been having issues as well. At least most of the issues with AMD have been resolved in the latest driver. I guess if D3D 9 mode is causing issues just used D3D 11 mode :).

[ryrynz]

How about Jinc SR4 vs NGU Very High for Chroma?

Jinc SR4 vs NGU Very High (http://screenshotcomparison.com/comparison/193277)

[ryrynz]

chroma is way overrated



You will not convince me to waste gpu and use anything other than Bicubic 60AR



I see zero difference in real world viewing.


Let's try this using a similar scene then but larger, from the film Kubo and the Two Strings.

Bicubic 60 AR vs NGU Very High. (http://screenshotcomparison.com/comparison/193279)

[AngelGraves13]

Jinc SR4 vs NGU Very High (http://screenshotcomparison.com/comparison/193277)

NGU looks pretty nice for Chroma...

One last request...

NGU Very High vs NGU Very High SR4

I want to see if SuperRes helps at all with Chroma for NGU

[ryrynz]

I want to see if SuperRes helps at all with Chroma for NGU

It certainly makes things brighter. Is it better? Who knows, up to you to decide.

NGU Very High vs NGU Very High SR4 (http://screenshotcomparison.com/comparison/193284)

[CruNcher]

Let's try this using a similar scene then but larger, from the film Kubo and the Two Strings.

Bicubic 60 AR vs NGU Very High. (http://screenshotcomparison.com/comparison/193279)


Nice example i guess everyone can percept the shaprness difference here like in Neo-XP his examples with NGU-Med :)

Your example becomes even more fascinating because of the lower filesize of the better perceptable psy result.

I mean we talking in your example about almost 10kb less for this particular frame with a overall better psy result.

But this was allready visible in Neo-XP his examples though not really with a final filesize reduction but the opposite.

So your example is kinda also crazy in it's overall result there seems to be a compression benefit as well now ?.

We definitely need some better more scientific analysis of what madshi does here ;)

[AngelGraves13]

It certainly makes things brighter. Is it better? Who knows, up to you to decide.

NGU Very High vs NGU Very High SR4 (http://screenshotcomparison.com/comparison/193284)

I see an improvement. The colors pop out more and the borders are thicker.

[huhn]

more red on black comparison.

bicubic 60 AR: https://abload.de/img/bicubic60arq0ltz.png
bilateral sharp:https://abload.de/img/bilateralsharpuza5g.png
NGU low: https://abload.de/img/ngulowwbzq1.png
NGU very high: https://abload.de/img/nguveryhighqglpe.png
recon sharp: https://abload.de/img/reconstructionsharprjbmr.png
super XBR 100 AR: https://abload.de/img/superxbr100ar43a0c.png
super XBR 150 AR: https://abload.de/img/superxbr150arkra6a.png

super XBR 150 AR is doing an fantastic job for it's price on this image and bilateral is doing pretty bad here and that is the most obvious thing to "learn" here.

@StinDaWg

i'm pretty sure he got your point.
but the chroma scaler aren't there to waste electricity. people created these scaler and put a lot of work in them and they are not joking around.

if you don't see the difference FINE! but why do you have to stop them comparing chroma scaler?

@khanmein
nothing changed for madVR usages so whatever:
http://bugs.madshi.net/view.php?id=431

[StinDaWg]

NGU doubling is really impressive with 720p animation @ 1080p. After playing back an episode of Family Guy in 720p and then 1080p version, it was hard to tell the difference. It's easy to tell the difference with Jinc and NNEDI3, 720p version looks blurry. Switch to NGU and bam, back in focus. The only thing that stands out is aliasing on some diagonal lines with NGU, otherwise they look almost identical.

if you don't see the difference FINE! but why do you have to stop them comparing chroma scaler?
I'm not stopping anyone from doing anything. The differences to me are not noticeable in 99% of real world circumstances. All these comparisons of red text are not applicable to what you see watching in the majority of a video. I still haven't come across anything where I paused a video, changed chroma, then saw any difference at all. With cartoons, movies, sports or anything else.

But, if the only way to tell the difference is with these extreme fringe examples, then more power to them. That will be my last comment on the matter. It really doesn't matter to me what other people use. I'm just glad something like madVR exists for us videophiles. :)

[Neo-XP]

more red on black comparison.

bicubic 60 AR: https://abload.de/img/bicubic60arq0ltz.png
bilateral sharp:https://abload.de/img/bilateralsharpuza5g.png
NGU low: https://abload.de/img/ngulowwbzq1.png
NGU very high: https://abload.de/img/nguveryhighqglpe.png
recon sharp: https://abload.de/img/reconstructionsharprjbmr.png
super XBR 100 AR: https://abload.de/img/superxbr100ar43a0c.png
super XBR 150 AR: https://abload.de/img/superxbr150arkra6a.png

super XBR 150 AR is doing an fantastic job for it's price on this image and bilateral is doing pretty bad here and that is the most obvious thing to "learn" here.

@StinDaWg

i'm pretty sure he got your point.
but the chroma scaler aren't there to waste electricity. people created these scaler and put a lot of work in them and they are not joking around.

if you don't see the difference FINE! but why do you have to stop them comparing chroma scaler?

@khanmein
nothing changed for madVR usages so whatever:
http://bugs.madshi.net/view.php?id=431

Yes, be careful with Bilateral, it can give you the best, but also the worst results depending on the situation.

Super-xbr is much more polyvalent, with values of 100-150.
I use NGU-Med at the moment (or NGU-High on new versions), because it is overall sharper and cleaner... and it is cold now where I live :)

Example with BilateralSharp / Bicubic60 AR / super-xbr150 / NGU-Low / NGU-Med / NGU-High :

https://s29.postimg.org/46stvbn77/0_Bilateral_Sharp.jpg (https://postimg.org/image/46stvbn77/) https://s29.postimg.org/3vbdik6r7/1_Bicubic60.jpg (https://postimg.org/image/3vbdik6r7/) https://s29.postimg.org/otm4udvzn/2_Super_xbr150.jpg (https://postimg.org/image/otm4udvzn/) https://s29.postimg.org/ct0otnokz/3_NGU_Low.jpg (https://postimg.org/image/ct0otnokz/) https://s29.postimg.org/a07h9mo8j/4_NGU_Med.jpg (https://postimg.org/image/a07h9mo8j/) https://s29.postimg.org/jm11pxfeb/5_NGU_High.jpg (https://postimg.org/image/jm11pxfeb/)

I zoom for you in the right spot :

https://s24.postimg.org/6jm6nwr29/0_Bilateral_Sharp.png (https://postimg.org/image/6jm6nwr29/) https://s24.postimg.org/mvw8dn5dt/1_Bicubic60.png (https://postimg.org/image/mvw8dn5dt/) https://s24.postimg.org/w4yenre9t/2_Super_xbr150.png (https://postimg.org/image/w4yenre9t/) https://s24.postimg.org/58efffvgh/3_NGU_Low.png (https://postimg.org/image/58efffvgh/) https://s24.postimg.org/ufpbfoykh/4_NGU_Med.png (https://postimg.org/image/ufpbfoykh/) https://s24.postimg.org/v6i1lh0xt/5_NGU_High.png (https://postimg.org/image/v6i1lh0xt/)

Now I let you decide which one to use.

[StinDaWg]

more red on black comparison.

bicubic 60 AR: https://abload.de/img/bicubic60arq0ltz.png
bilateral sharp:https://abload.de/img/bilateralsharpuza5g.png
NGU low: https://abload.de/img/ngulowwbzq1.png
NGU very high: https://abload.de/img/nguveryhighqglpe.png
recon sharp: https://abload.de/img/reconstructionsharprjbmr.png
super XBR 100 AR: https://abload.de/img/superxbr100ar43a0c.png
super XBR 150 AR: https://abload.de/img/superxbr150arkra6a.png

super XBR 150 AR is doing an fantastic job for it's price on this image and bilateral is doing pretty bad here and that is the most obvious thing to "learn" here.
I don't see any difference between these images. What am I supposed to be looking for? I guess it's possible that either my plasma or Firefox is doing something that is negating the effects, because they all look the same on this tv.

[huhn]

try to compare bilateral and super XBR 150.

most obviously the red triangles but not limited to them.

i don't think it is possible to hide the differences with a sub sampling screen. maybe vivid mode or something like that can hide this but...

i'm pretty sure firefox can't hide this what so ever.

this is not a chroma heavy image but it still shows a huge difference with bilateral.

[StinDaWg]

OK. I see a slight difference in sharpness/clarity with the red triangles between bilateral and super XBR 150. Less so with bicubic 60AR. If that's all I'm missing by using a faster algorithm, I'm fine with that. There's no way I could see that in motion anyways.

[Zaoshi]

Hello, I seem to be having some weird performance behaviour with NGU.

In one case madVR v0.91.3 NGU very high was taking 90 ms, but if I switch to NNEDI > Nearest > NGU very high it goes down to 16 ms for a dozen seconds and then back to 90 ms.: https://www.youtube.com/watch?v=Vt8FdGAgXL0

On madVR v0.91.4 similar situation has occurred on a different video. Using NGU high I watched few episodes with 30 ms / frame performance, but then it suddenly went up to 175 ms / frame. I tried previous trick and it went down to 40 ms / frame. While I was writing this post it has managed to go back to 30 ms / frame.: https://www.youtube.com/watch?v=tRxcFWl4qfA
Edit: it refuses to play at 30 ms / frame anymore. Had to revert back to NNEDI to continue watching.
Edit2: one episode later NGU works properly again.
Edit3: I think I figured it out. My GPU goes to sleep for some reason and reduces performance to save power and lower temperature. Any idea how to prevent this?

Not sure whether this is madVR issue or something with my system, but this never happened back when I was using NNEDI.

I have GeForce GTX 1070 with driver version 376.19.

[oddball]

I'm having a couple of issues with fast forwarding/rewinding on my bluray remuxes. The screen flickers on and off for a while and I get screen corruption until it seems to catch up on itself. No such issues with encodes. The other issue I have is if I put 1080p23 or 1080p24 in my OLED TV's display modes only half the video displays on the TV. I have to leave out those entries for MadVR to play them back correctly in fullscreen exclusive mode on the TV.

Any ideas?

[huhn]

OK. I see a slight difference in sharpness/clarity with the red triangles between bilateral and super XBR 150. Less so with bicubic 60AR. If that's all I'm missing by using a faster algorithm, I'm fine with that. There's no way I could see that in motion anyways.

well on my screen the difference between bilateral and super XBR 150 is is way bigger than luma FHD -> UHD scaling with bilinear or NGU at least on that source.

[StinDaWg]

well on my screen the difference between bilateral and super XBR 150 is is way bigger than luma FHD -> UHD scaling with bilinear or NGU at least on that source.
Ya, it's noticeable, but I wouldn't use bilinear anyways.

[ashlar42]

Ys, be careful with Bilateral, it can give you the best, but also the worst results depending on the situation.e

Super-xbr is much more polyvalent, with values of 100-150.
I use NGU-Med at the moment (or NGU-High on new versions), because it is overall sharper and cleaner... and it is cold now where I live :)

Example with BilateralSharp / Bicubic60 AR / super-xbr150 / NGU-Low / NGU-Med / NGU-High :

https://s29.postimg.org/46stvbn77/0_Bilateral_Sharp.jpg (https://postimg.org/image/46stvbn77/) https://s29.postimg.org/3vbdik6r7/1_Bicubic60.jpg (https://postimg.org/image/3vbdik6r7/) https://s29.postimg.org/otm4udvzn/2_Super_xbr150.jpg (https://postimg.org/image/otm4udvzn/) https://s29.postimg.org/ct0otnokz/3_NGU_Low.jpg (https://postimg.org/image/ct0otnokz/) https://s29.postimg.org/a07h9mo8j/4_NGU_Med.jpg (https://postimg.org/image/a07h9mo8j/) https://s29.postimg.org/jm11pxfeb/5_NGU_High.jpg (https://postimg.org/image/jm11pxfeb/)

I zoom for you in the right spot :

https://s24.postimg.org/6jm6nwr29/0_Bilateral_Sharp.png (https://postimg.org/image/6jm6nwr29/) https://s24.postimg.org/mvw8dn5dt/1_Bicubic60.png (https://postimg.org/image/mvw8dn5dt/) https://s24.postimg.org/w4yenre9t/2_Super_xbr150.png (https://postimg.org/image/w4yenre9t/) https://s24.postimg.org/58efffvgh/3_NGU_Low.png (https://postimg.org/image/58efffvgh/) https://s24.postimg.org/ufpbfoykh/4_NGU_Med.png (https://postimg.org/image/ufpbfoykh/) https://s24.postimg.org/v6i1lh0xt/5_NGU_High.png (https://postimg.org/image/v6i1lh0xt/)

Now I let you decide which one to use.
I don't understand why you guys go to such great length to provide comparisons shot and don't include the original. How is one supposed to judge anything like this?

[huhn]

well because it is not so easy to share a 4:2:0 screen.

[Neo-XP]

I don't understand why you guys go to such great length to provide comparisons shot and don't include the original. How is one supposed to judge anything like this?

Here is the same frame saved by MPC before processing :

https://s29.postimg.org/90b2lfab7/original.png (https://postimg.org/image/90b2lfab7/)

It is from a movie, so I hope this is what you want. If not, tell me how to extract it :thanks:

Or just pick the one which looks more natural or which you like the best based on... anything you want. You decide (but don't pick Bilateral ! :p).

[ashlar42]

well because it is not so easy to share a 4:2:0 screen.It might be, but without the original picture I still don't understand what you are evaluating. The upscaled picture you like the most based on...? What?
I'm not trolling here, I'm just genuinely curious.

[huhn]

well i'm not saying you are trolling at all.

it is just adding a 4:2:0 picture is a quite some work.

i usually add the source image when i'm doing image scaling comparison it is simply to much work for chroma scaling. i mean getting the except frame and encode it lossless with for example x264 is not something every person can do.

[CruNcher]

i would really love to see a compare with Intels Hardware Adaptive Lanczos 4 tap (Intel HD) but it's not easy todo at all.

[huhn]

why? just select DXVA scaling and use an intel iGPU.

it is more like lanczos 3 AR with banding thanks to low bit deep math.