madVR - high quality video renderer (GPU assisted)

[huhn]

madVR doesn't do any unnecessary processing by default.

chroma is upscaled everytime and this is unavoidable.

[Asmodian]

should I be using madvr if I only ever watch 1080p and 4k, but I need HDR passthrough to TV

Yes. There are some steps that must be done to get the video from the file to your TV and madVR does them in higher quality than anything else.

Chroma upscaling
Conversation to RGB
Dithering to the output bitdepth

All of these steps must happen to get 4K video on a 4K TV.

[ryrynz]

Also madvr's HDR tone mapping is set to get a boost in quality with the next version, likely surpassing many HDR television manufacturers own algorithms. So yes you will want to use madVR for 4K HDR.

[mytbyte]

I couldn't find this covered before so here it is: I'm trying to calibrate a SDR UHD TV to MadVR HDR->SDR conversion by using HCFR and rendering via MADTPG to avoid advancing the HDR patterns manually from disk but it seems MADTPG outputs regular gamma style curve i.e. patterns are not HDR encoded, even though I have HDR->SDR conversion activated in MadVR settings. Is it possible that MADTPG doesn't work with MadVR's own HDR processing options but only when a HDR TV is connected? Are there plans to allow MADTPG generate HDR patterns without a HDR display connected and pass them through HDR->SDR processing? Madshi?

[huhn]

HDRT -> SDR conversation is a conversation to "gamma" there is no point in sending HDR pattern and there is no need for a different calibration if your SDR is calibrated.

[mytbyte]

HDRT -> SDR conversation is a conversation to "gamma" there is no point in sending HDR pattern and there is no need for a different calibration if your SDR is calibrated.

You have the point...in theory. But my SDR gamma is pretty much spot on 2.2 all the way but the measures from manual Masciola's patterns don't track as close to BT.2390 roll off as I'd like so I like to tweak it more while looking at the BT.2390 curve which would be much easier going back and forth with automatic patterns.

[Asmodian]

I am not sure what you mean by "patterns are not HDR encoded", do you mean they do not go through madVR's HDR -> SDR conversion path? How would madTPG know HCFR was sending patches it should treat as HDR? I don't think madTPG looks for flags from software telling it HDR metadata, I don't think there are any standards around metadata attached to test patches so it would have to be something worked out between HCFR (Argyllcms) and madVR.

An option in madTPG to set the input HDR metadata, similar to how we can set the output metadata, would be cool. :)

[huhn]

HDR pattern without HDR are pointless.

[Manni]

I am not sure what you mean by "patterns are not HDR encoded", do you mean they do not go through madVR's HDR -> SDR conversion path? How would madTPG know HCFR was sending patches it should treat as HDR? I don't think madTPG looks for flags from software telling it HDR metadata, I don't think there are any standards around metadata attached to test patches so it would have to be something worked out between HCFR (Argyllcms) and madVR.

An option in madTPG to set the input HDR metadata, similar to how we can set the output metadata, would be cool. :)

MadTPG suports HDR, at least with Calman. You simply have to specify the metadata for the content (max brightness etc) and the software indicates this to MadTPG and switches to HDR. This has been the case for a while. It’s only useful if calibrating in HDR (passthrough), not if you’re doing a conversion, in which case you want the display to be calibrated to sdr rec-709 or P3 or BT2020 with a power gamma (2.4 recommended) and you don’t need/want MadTPG to be in HDR mode.

HCFR should be supporting this, best would be to ask Zoyd on AVS in the HCFR thread.

[mytbyte]

I am not sure what you mean by "patterns are not HDR encoded", do you mean they do not go through madVR's HDR -> SDR conversion path? How would madTPG know HCFR was sending patches it should treat as HDR? I don't think madTPG looks for flags from software telling it HDR metadata, I don't think there are any standards around metadata attached to test patches so it would have to be something worked out between HCFR (Argyllcms) and madVR.

An option in madTPG to set the input HDR metadata, similar to how we can set the output metadata, would be cool. :)

Your text in bold, that's what I meant and thought was implemented from the go since it comes from the same author. Viewing the PQ curve tells the whole story, can't tell how close it actually tracks from SDR luminance curve measures.

MadTPG suports HDR, at least with Calman. You simply have to specify the metadata for the content (max brightness etc) and the software indicates this to MadTPG and switches to HDR. This has been the case for a while. It’s only useful if calibrating in HDR (passthrough), not if you’re doing a conversion, in which case you want the display to be calibrated to sdr rec-709 or P3 or BT2020 with a power gamma (2.4 recommended) and you don’t need/want MadTPG to be in HDR mode.

HCFR should be supporting this, best would be to ask Zoyd on AVS in the HCFR thread.

I asked, no reply so far. Why do you think 2.4 gamma is recommended, i.e. why does it matter any if we can select the gamma a display is calibrated to in MadVR?

[huhn]

if your PQ curve is incorrect with a proper SDR calibration than the calibration is not the problem it'S the conversation.

so changing the calibration to get better PQ is the wrong way to look at it what you should look at is the conversation.

[mytbyte]

if your PQ curve is incorrect with a proper SDR calibration than the calibration is not the problem it'S the conversation.

so changing the calibration to get better PQ is the wrong way to look at it what you should look at is the conversation.

OK. So you think MadVR's implementation of BT.2390 may be wrong? I found the previous implementation of fixing 100 nits as difuse white didn't quite work. (far from bashing madshi, he's doing a devine job)

[Manni]

I asked, no reply so far. Why do you think 2.4 gamma is recommended, i.e. why does it matter any if we can select the gamma a display is calibrated to in MadVR?

You can pick any gamma you want with MadVR, but with a standalone player gamma 2.4 is recommended as it’s the likely target for the conversion.

[huhn]

OK. So you think MadVR's implementation of BT.2390 may be wrong? I found the previous implementation of fixing 100 nits as difuse white didn't quite work. (far from bashing madshi, he's doing a devine job)

there is far more than one type of HDR conversation in madVR and most important it is not a "simple" bt 2390 conversation.

[mytbyte]

there is far more than one type of HDR conversation in madVR and most important it is not a "simple" bt 2390 conversation.

@huhn: BT.2390 standardizes the mapping AFAIK, and there are only 2 more options to choose from for tone mapping, "clipping" and "arve curve", the latter I'm not familiar with...and I really don't like the cliping approach even if the rest of the curve is prefect

[huhn]

the "fix too bright..." will have some effect on the gamma curve.
i don't see a way to dynamical reduce luminance without changing the gamma. but obviously not sure how it works.

[Polopretress]

Hello.
I am not sure it is the best place to discuss on evolution request and if there is a more situable/dedicated area, please let me know.

The subject is linked to the syntax used in the profile rules to execute an external program when a profile is activated.
The goal is to launch a programm depending of the aspect ratio AR of the movie.
This program will modify the position of the mask of the screen to fit the aspect ratio of the movie.

Today , with the actual behaviour, is it possible to add a variable in argument with the program to execute. This variable would be , of course, the value of the numerical parameter AR

If it is possible, what is the correct syntax ? for example the line command to execute could be : c:\mask.exe AR or c:\mask.exe Value(AR)
If it not possible today, is there a possibility to implement such behaviour to be able to use some numerical value managed by madVR into an external program ?

[mytbyte]

the "fix too bright..." will have some effect on the gamma curve.
i don't see a way to dynamical reduce luminance without changing the gamma. but obviously not sure how it works.

Will check this out...I mean, what I'm already getting is probably better than many "HDR" TVs...

[huhn]

it'S clearly intentional and more improvements are planned.

[veggav]

Today I took a few hours, the Godfather Blu-ray and played it on the OPPO 103 and MadVR.

I tried every upscaling shader and ended up using NGU AA very high.

Still, the OPPO seemed better with less noise.
But I think I found a way to make both images identical, there's a noise reduction setting on nvidia control panel, if you set it to 30~50% the sharpned grain by MadVR becomes less noticeable and the noise patterns are almost identical.

I know the newest MadVR have denoise feature but since I'm using Jriver and I use BD and UHD BD menus I can't update.
Tried MadMax UHD recently with the newest MadVR and JRiver and there were subtitles only if the pop up menu was open.
(not blaming MadVR, just commenting).

I would like to ask you, since MadVR is doing all the upscaling do you guys use Cinema mode with your display?
I believe Game or Graphics/PC mode would give you a more raw image but since MadVR is dealing with it, is there a point in using it?

[huhn]

the video settings from nvidia are not supposed to work with madVR outside of DXVA scaling.

[veggav]

Hum.. maybe placebo effect? It's possible it's was a few hours going back and forth.

[Asmodian]

I believe Game or Graphics/PC mode would give you a more raw image but since MadVR is dealing with it, is there a point in using it?

Because madVR is dealing with it you want to use a PC more more. Using Cinema mode destroys much of the quality benefit of madVR's high quality chroma upscaling and dithering. Your display gets the data after madVR.

(this is assuming Cinema mode does its processing in 4:2:2 while PC is 4:4:4 or RGB, which is common)

[Warner306]

@huhn: BT.2390 standardizes the mapping AFAIK, and there are only 2 more options to choose from for tone mapping, "clipping" and "arve curve", the latter I'm not familiar with...and I really don't like the cliping approach even if the rest of the curve is prefect

The BT.2390 curve is converted to an SDR gamma curve, so you are really attempting to track an SDR gamma curve with an input that is too bright.

[Warner306]

Hum.. maybe placebo effect? It's possible it's was a few hours going back and forth.

Noise with madVR's upscaling has been mentioned a few times recently. It is surprising the Oppo is less sharp or it is using some type of denoising filter. The HTPC shouldn't be creating noise not found in the source. Have you tried sending a limited range signal instead of a full range signal to see if your display is doing a poor RGB conversion? Combining NGU Sharp with the free version of reduce compression artifacts can sometimes help but will soften titles with film grain.

[veggav]

Because madVR is dealing with it you want to use a PC more more. Using Cinema mode destroys much of the quality benefit of madVR's high quality chroma upscaling and dithering. Your display gets the data after madVR.

(this is assuming Cinema mode does its processing in 4:2:2 while PC is 4:4:4 or RGB, which is common)

The text is way more visible in graphics mode compared to cinema but I'm not sure if chroma subsampling is happening here.
On my previews TV when 4:2:2 mode were on I could barely see the text, so I'm not enterily sure.

https://ibb.co/ges2n8
https://ibb.co/cYvcLT

crap cellphone pictures from the screen but I believe it's possible to have an idea.

[huhn]

TV with heavy sharpening can kind of hide chroma subsampling on test pattern.

this is pretty obvious.

[Asmodian]

ges2n8 looks like 4:2:2 to me.

[veggav]

TV with heavy sharpening can kind of hide chroma subsampling on test pattern.

this is pretty obvious.

The sharpness goes from 0 to 100, it's on default position here: 50.
That's what is recommended by rtings review of the Sony 900E.

Actually, setting sharpness is quite dificult to do with 4k because you can't spot on the right point where noise starts to be added.

I'll stick with graphics/game mode from now on.

[huhn]

sony neutral is usually at 50%.

and no offence but i have a hard time to believe/trust your noise claims.

[ryrynz]

Actually, setting sharpness is quite difficult to do with 4k because you can't spot on the right point where noise starts to be added.


You're not looking for noise, you're looking for ringing / changes in brightness.

[veggav]

Noise with madVR's upscaling has been mentioned a few times recently. It is surprising the Oppo is less sharp or it is using some type of denoising filter. The HTPC shouldn't be creating noise not found in the source. Have you tried sending a limited range signal instead of a full range signal to see if your display is doing a poor RGB conversion? Combining NGU Sharp with the free version of reduce compression artifacts can sometimes help but will soften titles with film grain.

Hi Warner, I'll test limited range tonight and see if that makes a difference.

When you are referring to reduce compression artifacts you mean a separate software or an option in a recent version of MadVR?
I remember a denoise option in the newest Madvr but I can't use it with Jriver and UHD BD menus for now.

By the way, what's the consensus on the quality of DXVA scalling? Terrible? Worth looking into?

[Warner306]

Hi Warner, I'll test limited range tonight and see if that makes a difference.

When you are referring to reduce compression artifacts you mean a separate software or an option in a recent version of MadVR?
I remember a denoise option in the newest Madvr but I can't use it with Jriver and UHD BD menus for now.

By the way, what's the consensus on the quality of DXVA scalling? Terrible? Worth looking into?

If you choose NGU Sharp, you can choose reduce compression artifacts under artifact removal with a checkbox that adds it for no additional resource use.

You might as well just use Kodi VideoPlayer if you are going to use DXVA scaling. It is just hardware scaling and not that great if you are using a Nvidia or AMD GPU.

[mrmojo666]

If you choose NGU Sharp, you can choose reduce compression artifacts under artifact removal with a checkbox that adds it for no additional resource use.

You might as well just use Kodi VideoPlayer if you are going to use DXVA scaling. It is just hardware scaling and not that great if you are using a Nvidia or AMD GPU.

just an opinion about dxva upscaling in madvr, when upscale 720/1080 to 4k i notice that it is way better than using internal renderer of mediaportal or mpc-hc. maybe worse than ngu, honestly on my tv (55") the electricity used with ngu doesn't worth the expense. With < 720 I can see ngu is totally another universe than dxva.

[huhn]

than just use bicubic in madVR at least you don't get banding or wrong position chroma this way.

the EVR renderer in MPC-HC is using DXVA scaling. by default EVR-CP is using bilinear you can change this to bicubic which is again similar to AMD NVIDIA DXVA scaling.

[waldnebel]

Why are the rendering stats of the last 5 seconds in the fse mode this different from the average stats?


What I get as information is the following:


D3D11 exclusive (10 bit)

average stats
rendering: 26.0ms


max stats (5s)
rendering: varying from 26.0ms to mainly 46.0ms


What is causing this strange 46.0ms peak? Dropped or repeated frames are not reported.


If I activate that windowed overlay mode I get a constant rendering time of around 21.0ms on both sides but that mode gets identified as D3D11 fullscreen windowed (8 bit) and that 8 bit is causing a lot of headaches. I was expecting the same as in the fse mode.

I'm using Windows 10, Version 1803 and the nvidia driver 398.36. The power management mode is set to Optimal power. The CPU is set to High Performance.


I have probably also a last question, then I'm done configuring madvr. Should I leave the gamma correction slider of my TV alone?


Thank you for that clarification regarding the HDR option of windows 10, Asmodian.

[SirSwede]

Today I took a few hours, the Godfather Blu-ray and played it on the OPPO 103 and MadVR.

I tried every upscaling shader and ended up using NGU AA very high.

Still, the OPPO seemed better with less noise.
But I think I found a way to make both images identical, there's a noise reduction setting on nvidia control panel, if you set it to 30~50% the sharpned grain by MadVR becomes less noticeable and the noise patterns are almost identical.

I know the newest MadVR have denoise feature but since I'm using Jriver and I use BD and UHD BD menus I can't update.
Tried MadMax UHD recently with the newest MadVR and JRiver and there were subtitles only if the pop up menu was open.
(not blaming MadVR, just commenting).

I would like to ask you, since MadVR is doing all the upscaling do you guys use Cinema mode with your display?
I believe Game or Graphics/PC mode would give you a more raw image but since MadVR is dealing with it, is there a point in using it?

I have found using Madvr in Game Mode produces the best picture quality.

[Asmodian]

The power management mode is set to Optimal power.

Do not use optimal power, that is too conservative. Try adaptive.

A few spikes up to 46ms should not be a problem, that is what the buffers are for.

To use windowed overlay you must turn off D3D11, D3D11 overrides other options for rendering API. Why is 8-bit a problem? 8 bit is often higher quality than 10 bit (like on my 2017 LG OLED) and 10 bit is not very important even when the display handles it well. Can you tell the difference between?

Gamma should be correct on your TV, setting it on the TV would normally be part of calibration, which I would do before a software calibration or configuring madVR. You can adjust gamma in madVR but it is usually better to do it on the TV if you can.

[sauma144]

@madshi
I know you are working hard on HDR stuffs but can we expect other improvements or features in the next release like a sharper NGU Sharp, anime optimized NGU or maybe new upscalers for lowend/fanless/laptop devices?
Is it true your new algo will be a SVP killer? :rolleyes:

[el Filou]

A *sharper* NGU Sharp? It's amazingly sharp already. :confused:

[ryrynz]

@madshi
I know you are working hard on HDR stuffs but can we expect other improvements

There will be other improvements in time, but for the time being.. nope to everything you said :P

[sauma144]

A *sharper* NGU Sharp? It's amazingly sharp already. :confused:
It was an example, I don't know if it's possible to make it sharper without destroying the picture.

[LigH]

Of course not. "Sharpness" can only exist in an original picture. Every other attempt to filter a soft image only increases contrasts, but restoring or even creating originally lost sharpness is impossible. For reasons, see Wikipedia: Gibbs phenomenon (https://en.wikipedia.org/wiki/Gibbs_phenomenon), plus try to imagine how to discover which summands were added to a sum, how many valid results this question may have (as results of any scaling are all weighted sums). The best one can try is limiting annoying side effects like the mentioned "ringing artifacts", but the result may still be an artificial "plastic" look.

[Chyrka]

Hello.
Not specifically related to madvr, but maybe I will get answer without creation of separate topic.
I failed to find any similar issue with google and forum search.

I have a problem with chroma shift 0.5 pixel to right and 0.5 pixel to bottom on any 4:2:0 video any player any renderer when my monitor is connected to my NVIDIA 1060 (driver 388.59, not the freshest at the moment).
At the same time my old integrated Intel HD3000 doesn't have such an issue.
Here are some illustrations for horizontal shift:
V plane, taken in avspmod, doubled in size:
1) as it is in the original video
2) as it is shown by Intel graphics
3) as it is shown by Nvidia graphics
http://thumbs2.imagebam.com/ff/dd/3b/a7b5c5908666754.jpg (http://www.imagebam.com/image/a7b5c5908666754) http://thumbs2.imagebam.com/74/1f/59/bf274d908666764.jpg (http://www.imagebam.com/image/bf274d908666764)
Ultrazoomed example
https://puu.sh/AOK8y.png
https://puu.sh/AOKdl.png
Win7, Version of madvr 0.91.11, default chroma scaling but I think it doesn't matter as far as it is the same issue for other programs.
Please help or say where should I go. Thanks.

[Asmodian]

Are you using any DXVA scaling options? This is a known issues with Nvidia's DXVA scaling.

[Warner306]

Changing the video decoding method might also help.

[Chyrka]

Changing the video decoding method might also help.
Switching to DXVA (copy-back) in LAV decoder fixed the issue.
Thank you, guys Warner306, Asmodian.

[Warner306]

What were you using before?

[Chyrka]

What were you using before?
DXVA2 (native) - the default value, if I am not mistaken.

[foozoor]

Can we expect other improvements or features in the next release like a sharper NGU Sharp, anime optimized NGU or maybe new upscalers for lowend/fanless/laptop devices?
Is it true your new algo will be a SVP killer?
FSRCNNX is already sharper than NGU-AA, the SSIM variant is even more sharper.
Everybody knows the madshi's new algo is a motion interpolation one.