madVR - high quality video renderer (GPU assisted)

[Asmodian]

No, that is fine. Maybe you need to reset to default settings first?

Very odd, sorry.

[steakhutzeee]

So, i've:

Chroma Upscaling -> NGU Sharp, medium. No SuperRes (MadVr stops me only if i use NGU and SuperRes from the upscaling refinement, but'll follow your suggestion).

Image Downscaling -> SSIM 1D 100%, No in linear light, Yes anti-ringing relaxed, and Yes anti-bloating 100%. What's difference with 2D? What i should use?

Image Upscaling -> NGU Sharp, High for luma doubling, all the others are let madVR decide.

Upscaling Refinement -> I've sharpen edges and enhance detail because as i can see, the first sharpens only the edges, the second sharpens the texture etc. I need them? I've enabled crispen edges, but it's another sharpener. So i need it? Disabled soften edges.
Enabled LumaSharpen 0.65, to avoid sharpen artifacts.
Enabled AdaptiveSharpen 0.5, to avoids sharpening near flat areas and very sharp edges. (Asmodian guide).
No SuperRes, and checked refine image after every 2x upscaling step.

Thanks in advance.

[Klaus1189]

Originally Posted by Klaus1189

I'm trying to let madvr measure video playback. But it is still displaying "not measured yet" in this video mode. I followed the four steps. What am I doing wrong?

So I don‘t get optimization data. What could prevent this?

[nussman]

What do you think? Did you test your settings? I dont think Lumasharpen and AdaptiveSharpen helps for most content but there is no "best" setting for videoprocessing. Test it and if you like it everything is fine ...

[Asmodian]

Originally Posted by steakhutzeee

Image Downscaling -> SSIM 1D 100%, No in linear light, Yes anti-ringing relaxed, and Yes anti-bloating 100%. What's difference with 2D? What i should use?

2D is an even stronger version of SSIM (more detail preserved and/or sharper) but it is perhaps too strong and it takes A LOT more GPU power.

You do not need any Upscaling Refinement options. I only use them for particular sources that are unusually soft or have some other reason to want them. By default I have them all disabled.

AdaptiveSharpen works by sharpening that way, it does not affect the other sharpeners at all.

Enabling them all is a bad idea, I have never found them to be useful all at once. Some sources like sharpening with crispen edges while others look better with sharpen edges or adaptive sharpen, they all have their place and sometimes a few look good together but just using them all by default is not good.

Originally Posted by Klaus1189

So I don‘t get optimization data. What could prevent this?

Usually this is because your GPU driver is not actually switching to the correct custom mode. Try having madVR do the switch instead of using the control panel, or vice-versa if you are already having madVR do the switch.

[steakhutzeee]

Originally Posted by Asmodian

2D is an even stronger version of SSIM (more detail preserved and/or sharper) but it is perhaps too strong and it takes A LOT more GPU power.

You do not need any Upscaling Refinement options. I only use them for particular sources that are unusually soft or have some other reason to want them. By default I have them all disabled.

AdaptiveSharpen works by sharpening that way, it does not affect the other sharpeners at all.

Enabling them all is a bad idea, I have never found them to be useful all at once. Some sources like sharpening with crispen edges while others look better with sharpen edges or adaptive sharpen, they all have their place and sometimes a few look good together but just using them all by default is not good.



Usually this is because your GPU driver is not actually switching to the correct custom mode. Try having madVR do the switch instead of using the control panel, or vice-versa if you are already having madVR do the switch.

Thanks So l've disabled all the upscaling refinement. Btw, I just seen a movie and with my previous settings it was looking great XD. Now, I know what the settings do thanks to your guide, but how I can go deeper, to know when a source need one of them?

Inviato dal mio ASUS_Z00LD utilizzando Tapatalk

[Asmodian]

Originally Posted by steakhutzeee

Now, I know what the settings do thanks to your guide, but how I can go deeper, to know when a source need one of them?

Pure looking at the video with and without them. There are no correct answers, only what looks good to you.

[Fabulist]

Hello madshi and thank you for your amazing work.

I would like to ask your opinion regarding color spaces and depth on a matter that has haunted me for awhile now:

If all sources are 8-bit (rarely 10-bit) 1080p YCbCr 4:2:0 and I have madVR scaling chroma with NGU Sharp very high quality, would it be better to have the TV (also 1080p) set to (through NVIDIA's Control Panel):

a) 8-bit or 10-bit or 12-bit (Deep Color) YCbCr 4:2:2 (Pure Direct* greyed out / off)
b) 8-bit YCbCr 4:4:4 (Pure Direct* can be used)
c) 8-bit RGB (Pure Direct* can be used)

*I am confused as to what Pure Direct does exactly, it should in theory allow for a less altered signal assuming it is RGB or YCbCr 4:4:4. It is not allowed in YCbCr 4:2:2 at any bit-depth, although I doubt it makes any difference considering the source's quality and madVR functionality.

I am receiving consistently different answers from various sources including my own research - but I assumed that you would know what is best in combination with madVR.

Does it even matter assuming the conditions mentioned above? What do you think is correct?

[bcec]

I just got around playing around with new custom timings, but these timings are engaged only when D3D11 windowed mode (8-bit). When D3D11 full screen exclusive (10-bit) enabled, custom modes are not engaged. Any ideas?

EDIT: I disabled D3D11 FSE, and instead using D3D11 windowed mode (10-bit), and custom modes are engaged now.

[Asmodian]

Originally Posted by Fabulist

a) 8-bit or 10-bit or 12-bit (Deep Color) YCbCr 4:2:2 (Pure Direct* greyed out / off)
b) 8-bit YCbCr 4:4:4 (Pure Direct* can be used)
c) 8-bit RGB (Pure Direct* can be used)

This has a well understood and non-controversial answer, at least in this thread.

Given those options it is always c) 8-bit RGB, even with high bit depth sources.

a) YCbCr 4:2:2 is particularly bad; madVR outputs RGB and you don't want the GPU downscaling your chroma to half the horizontal resolution! That loses 1/3 of the video information and with good dithering the bit depth is significantly less important than spatial resolution. Chroma upscaling with NGU Sharp very high is rather pointless if you then downscale it again with bilinear.

b) YCbCr 4:4:4 would be better than 4:2:2, but there is no reason to have the GPU do an extra conversion. Video data must be RGB for humans to view it as a color image so this is always the final format, if madVR does the conversion, the display does, or something inbetween.

Where do you see "Pure Direct"? It sounds good, you don't want any extra processing.

[JarrettH]

Originally Posted by steakhutzeee

Thanks So l've disabled all the upscaling refinement. Btw, I just seen a movie and with my previous settings it was looking great XD. Now, I know what the settings do thanks to your guide, but how I can go deeper, to know when a source need one of them?

Inviato dal mio ASUS_Z00LD utilizzando Tapatalk

With blu-ray material, SuperRes (upscaling refinement) with a value of 2-3 was popular with the Jinc luma upscaler.

NGU-Sharp has pretty much eliminated the use of upscaling refinement because it's so sharp. Not only sharp, but it reconstructs a larger picture faithfully.

Super-xbr as a doubler or NGU-Standard with direct quadrupling are both good for SD (DVD resolution).

[SamuriHL]

Originally Posted by Asmodian

This has a well understood and non-controversial answer, at least in this thread.

Given those options it is always c) 8-bit RGB, even with high bit depth sources.

a) YCbCr 4:2:2 is particularly bad; madVR outputs RGB and you don't want the GPU downscaling your chroma to half the horizontal resolution! That loses 1/3 of the video information and with good dithering the bit depth is significantly less important than spatial resolution. Chroma upscaling with NGU Sharp very high is rather pointless if you then downscale it again with bilinear.

b) YCbCr 4:4:4 would be better than 4:2:2, but there is no reason to have the GPU do an extra conversion. Video data must be RGB for humans to view it as a color image so this is always the final format, if madVR does the conversion, the display does, or something inbetween.

Where do you see "Pure Direct"? It sounds good, you don't want any extra processing.

I'll answer for them lol. Pure direct is a setting on Panasonic plasma TV's that enables 4:4:4 with no extra processing. It's something I was messing around with myself the other day and ran into my 3d issue which is solved. Pure direct is disabled in 3d mode. But for 2d using rgb full in madvr gpu and tv, pure direct is pretty sweet.

Sent from my Pixel XL using Tapatalk

[Siso]

Originally Posted by Siso

Is it good to use anti-bloating filter (100%) on image downscaling?

http://imgbox.com/YKDfHkUw

Anyone?

[nevcairiel]

Judge with your own eyes. Its the only measure that actually matters.

[uusuke]

TV: sony ex700
graphics card: msi rx480
blueskyfrc version: v2.81
madvr version: 0.92.2 and 0.92.3
win 10

usually average stat is 12ms ,if I use madvr 0.92.2 , every 5- 10 seconds will burst into 20 ms, and then dropped frames, and turn off blueskyfrc on the normal, some people have encountered the same problem?
(madvr 0.92.1 will not)

[huhn]

@SamuriHL

Panasonic pure direct mode and no extra processing well no...
you may get 4:4:4 with it but that's about it and the issues with it range from banding to 3:2 judder with active sharpening. in term of PC mode Panasonic has a lot to learn...

[Manni]

Originally Posted by Asmodian

This has a well understood and non-controversial answer, at least in this thread.

Given those options it is always c) 8-bit RGB, even with high bit depth sources.

a) YCbCr 4:2:2 is particularly bad; madVR outputs RGB and you don't want the GPU downscaling your chroma to half the horizontal resolution! That loses 1/3 of the video information and with good dithering the bit depth is significantly less important than spatial resolution. Chroma upscaling with NGU Sharp very high is rather pointless if you then downscale it again with bilinear.

b) YCbCr 4:4:4 would be better than 4:2:2, but there is no reason to have the GPU do an extra conversion. Video data must be RGB for humans to view it as a color image so this is always the final format, if madVR does the conversion, the display does, or something inbetween.

Where do you see "Pure Direct"? It sounds good, you don't want any extra processing.

Agreed with the above in general, but if the color/hue controls are active on the display when sending RGB, it usually means that the display internally converts RGB to YCbCr, applies the colors/hue control and converts back to RGB.

In that case sending YCbCr 4:4:4 from the GPU means one LESS conversion (well, the same number but it's done by the GPU rather than the display), not one more .

If the color/hue controls in the display are greyed out/inactive when you send RGB, then it means that there is no internal RGB>YCB>RGB conversion and sending RGB indeed means one less conversion.

[huhn]

but the RGB -> YCbCr conversation can be done in high bit deep in the screen the GPU is at best outputting 12 bit.

and who said you can't do that in RGB?

[Manni]

Originally Posted by huhn

but the RGB -> YCbCr conversation can be done in high bit deep in the screen the GPU is at best outputting 12 bit.

and who said you can't do that in RGB?

The color/hue controls can only work in YCbCr AFAIK.

[huhn]

doesn't make any sense.

they may do a lot of processing in YCbCr but that doesn't mean you can't color correct an image in RGB. you can do about everything color related with a 3D LUT and that is clearly working in RGB.