madVR - high quality video renderer (GPU assisted)

[mclingo]

ah right, there is nothing you can do on the TV itself to correct the scaling, have you tried this since replacing the card?

[glc650]

ah right, there is nothing you can do on the TV itself to correct the scaling, have you tried this since replacing the card?Tried what since replacing the card?

[mclingo]

have you looked into why your HTPC image isnt fitting on your TV correctly, is it a driver issue on your PC or does the TV had an odd quirk which means it cannot show the aspect ratio correctly?, I'm guessing you've already cycled through all the aspect ratios options on the TV as you've no doubt had it a number of years now.

[mclingo]

@MADSHI - I wonder if you could maybe create a benchmark tool for MADVR, be interesting for people to properly compare card stats.

[Warner306]

the brightness level of the none compressed parts match at gamma 2.2.

so if you want the correct brightness levels you set this screen is already calibrated to your screens response where unlike SDR madVR will automatic change your "gamma" for HDR sources to 2.2.

as i said before i'm not a friend of this inconsistent behaviour between HDR and SDR.

I know one user was getting what looked like correct HDR gamma tracking at 2.20 with madVR set to clipping. But I sent that person a black clipping pattern and they were clipping some black. So I'm not sure if 2.20 is "correctly" following the PQ curve in all circumstances.

None of the SDR gamma curves have equal response to the original PQ curve:

Image: Gamma 2.20 vs. PQ ST.2084 (https://i.postimg.cc/KvSys13K/Gamma-2-20-vs-ST-2084.jpg)

I would assume madVR would have to tell the display to flash the correct amount of voltage to get a PQ value from an SDR gamma curve and use all kinds of dithering to fill in all of the extra shades of gray required at the low end of the curve.

When tone mapping, changing the target nits also radically changes the perceived gamma, so there is more to the gamma response than just choosing 2.20 or 2.40. It is more customizable than that. As long as you can see reference black, you can make the image darker or brighter as required to get a smooth transition from dark to light. The original tone curve will be compressed either way.

I think there are some who seem to be getting an accurate image with both 2.20 and 2.40. And there are some that I know of that get some black crush when madVR is set to 2.20. You have to test black clipping or you may notice any crush.

[huhn]

the thing is actually pretty simple. when madVR tone maps to SDR it will map to 2.2 and there is nothing wrong with that you have to choice a gamma. the problem i have now start when you are using gamma processing HDR can go totally nuts and tries to reach "some" gamma and that's not a good thing. the whole gamma matching with the TV makes total sense but one totally normal setting in it will deny this even when properly used for SDR content and the reason for this is simple the point when madVR applys gamma correction is done at two different spot instead of doing it in one spot different for HDR and SDR.

so if you have a properly 3D LUT calibrated screen it doesn't matter if you switch between 2.4 or 2.2 it will not clip. that'S not my problem anyway here. and even if your screen starts clipping for what ever reason that doesn't mean the 2.2 target is incorrect.

[Warner306]

2.20 was simply too dark to watch, even if the display was set to 2.20. Near black detail matters a lot when watching HDR content because the image is being tone mapped. SDR content is currently mastered at 2.40 on the most popular mastering displays and then processed to 2.20 by the display. I found the same combination does still work with HDR, even if it isn't considered correct.

[huhn]

then you shouldn't have a problem with this:
move it to color & gamma by adding a HDR-SDR gamma processing that is ticked by default with pure gamma curve 2.20 and a warning to not change it.
this makes sure you can still properly gamma process SDR and don't ruin HDR with this option.

lying to madVR what your display is really calibrated to to get a different result just for HDR is not optimal.

using a gamma processing option to get mathematical incorrect results is totally fine by me it' the users choice even if it is just used for the same reason we use different gammas for SDR to make dark part better visible for example.

[tp4tissue]

2.20 was simply too dark to watch, even if the display was set to 2.20. Near black detail matters a lot when watching HDR content because the image is being tone mapped. SDR content is currently mastered at 2.40 on the most popular mastering displays and then processed to 2.20 by the display. I found the same combination does still work with HDR, even if it isn't considered correct.

could that not just be the display ?

which probe are you using. My i1d3 works well, but my spyder5 messes up dark tones and causes crush.

[ryrynz]

@MADSHI - I wonder if you could maybe create a benchmark tool for MADVR, be interesting for people to properly compare card stats.Been asked many multiple times over the years.

[tp4tissue]

Been asked many multiple times over the years.

Madshi either works on a tool for demo'n e-peen, OR, he makes Highlight recovery even better..

Clearly, highlight recovery comes first.. :D

[ryrynz]

It's just as he's always said, buy the best card you can afford. ^_^ Performance wise it's Nvidia over AMD.

[mclingo]

where that is true, with amd cards being so cheap its hard to know how close they are price vs performance, do you get mid range nvidia or a high end amd, its not that simple anymore unless money is no object.

[tp4tissue]

It's just as he's always said, buy the best card you can afford. ^_^ Performance wise it's Nvidia over AMD.

I thought madshi ran ATI, what does Nvidia boards have that madvr makes better use of..

[Asmodian]

Something about NGU runs much better on Nvidia GPUs...

[tp4tissue]

where that is true, with amd cards being so cheap its hard to know how close they are price vs performance, do you get mid range nvidia or a high end amd, its not that simple anymore unless money is no object.

I just bought a used rx 580 $135, after the amd vega 7 announcement.

Since they haven't announced any mid-range product, there's nothing to hold out for this year.. Might as well get in on the Used Miner cards while they're still available.

[huhn]

the price differenve of a used 1060 and a used RX 580 is 30-40 bucks. and the performance difference using NGU and only NGU related scaler is about 100%-200% faster in favour of nvidia.

he said a year or longer ago that he had a 1070 and a 560 or 460.

pretty sure he has a RTX cards now to play around with the tensor core.

[tp4tissue]

the price differenve of a used 1060 and a used RX 580 is 30-40 bucks. and the performance difference using NGU and only NGU related scaler is about 100%-200% faster in favour of nvidia.

he said a year or longer ago that he had a 1070 and a 560 or 460.

pretty sure he has a RTX cards now to play around with the tensor core.

I have a 1060 in my kitchen pc..

The used 1060 on ebay looked slightly more raggedy than the used 580, so I went with 580.. This is for my basement pc.

[mclingo]

i'm guessing MADVR is generally more optimized for NVIDIA, it certainly feels that way and MADSHI has closer links with NVIDIA. It would be nice if he could take another look at polaris and see if he can do anything with it, given the number of AMD ex mining cards out there getting snapped up and great prices,

But I'm guessing his radar is already pretty full.

[huhn]

the market share of polaris cards is close to none existing and this will not change.
https://store.steampowered.com/hwsurvey/Steam-Hardware-Software-Survey-Welcome-to-Steam

and NGU runs great on none polaris card so how can he optimize for nvidia all run the same shader.

nnedi3 runs faster on AMD so he optimizes more for AMD? what is this reasoning...

[el Filou]

https://1drv.ms/u/s!AnsGKXR_EKR0hCfOm_6m0n5aXCFLThose presentation times are clearly an issue with the graphics driver.
If none of huhn's recommandations work, you can also try an older driver that still works with your newer card, or file a bug report with AMD.

Another trick: if you use your HTPC mostly for video, you can set your desktop at 720p and configure the media player to switch to 1080, then in the Radeon control panel you configure 720p with 3% underscan and 1080 with 0%. You'll still have the issue with GUI elements at the edges being cropped in MPC-HC, but it's a lesser problem than having the desktop GUI cropped IMHO.

[Warner306]

then you shouldn't have a problem with this:


lying to madVR what your display is really calibrated to to get a different result just for HDR is not optimal.

using a gamma processing option to get mathematical incorrect results is totally fine by me it' the users choice even if it is just used for the same reason we use different gammas for SDR to make dark part better visible for example.

I'm not sure going from 2.40 to 2.20 or 2.20 to 2.40 at the display is ruining HDR. Either way, it doesn't look half bad. I got the best results with 2.40. I can only relate my own experience.

[Warner306]

could that not just be the display ?

which probe are you using. My i1d3 works well, but my spyder5 messes up dark tones and causes crush.

I got similar results with two different displays when changing the gamma settings at the display and madVR. It could be some deficiency of the gamma tracking near-black, but who knows. They don't crush black with SDR content.

[chros]

I got the best results with 2.40. I can only relate my own experience.
Same for me, as I said before.
Yesterday I checked 2.4 gamma with black and white clipping of hdr10 test patterns (https://www.avsforum.com/forum/139-display-calibration/2943380-hdr10-test-patterns-set.html), again, and there's no issue at all.

[mclingo]

[QUOTE=huhn;1862500]the market share of polaris cards is close to none existing and this will not change.
https://store.steampowered.com/hwsurvey/Steam-Hardware-Software-Survey-Welcome-to-Steam

Yeah but thats steam, those figures are going to be massively skewed toward gamers, you cant call that "the market share", Most HTPC owners could own AMD cards, we just dont know.

NVIDIA cards arent just a bit faster according to some, they decimate AMD in NGU, surely further optimization is required, got to be a reason for this?

[huhn]

the reason should be an architectural difference that can't do the needed calculation fast enough or yet another driver bug.

NGU and polaris release time is very similar and it was very fast known that polaris doesn't perform well if i'm not mistaken i was one of the first if not the first to notice and investigate on this.

Yeah but thats steam, those figures are going to be massively skewed toward gamers, you cant call that "the market share", Most HTPC owners could own AMD cards, we just dont know.

not only was it hard to even get your hands on an polaris card when they where released the mining boom killed that product totally for the gaming market. if you now add the well known NGU problem it's not very great idea to get one if you want to use NGU.

fermi was decimate by GCN 1.0 for nnedi3 too this got better with kepler and was fixed with maxwell where AMD didn't even compete anymore and just refreshed old cards over and over.
and even then AMD was still faster and AMD had a bug where it lost about 50 % performance while decimating nvidia.
the 7850 and the r9 270 classics.

good old times. the fix not getting an nvidia card if you wanted nnedi3.

[madjock]

@Mclingo

1 Thread not enough :D

https://forum.kodi.tv/showthread.php?tid=223175&page=459

[DMU]

Does madVR have any API to display Overlay, like as Ctrl+J?

[huhn]

the subtitle interface or the api that shows the OSD was accessible too

[DMU]

Tell me, please, where can I get madVR's API for OSD?

[huhn]

sorry i'm not a programmer but it should be this: http://madshi.net/SubRenderIntf.h

[mclingo]

the reason should be an architectural difference tr9 270 classics.

good old times. the fix not getting an nvidia card if you wanted nnedi3.

i guess well have to hope AMD comes out with some new architecture at some point which works better, i'm happy with what my 580 can do for what i paid for it so i guess thats all that matters at the end of the day.

[Charky]

Same situation here but you dom't have to loose 12bits.
Do this: get CRU custom resolution utility.
Write down the front and back porch that madvr is using along with sync width.

Open cru, in detailed resolution add 3840x2160 23.000 with the information provided by madvr.
Click ok.
Press up so 3840x2160 23 will be right under 3840x2160 60.
Click ok
Use restart 64 or 32 depending on your system.

After the restart you will see 3840x2160 under PC resolution.

First under hd resolution, set 3840x2160 23 RGB Full dynamic range and 12bits for 23hz.
The custom resolution will be named 24hz.
Change it.
And now go to pc resolution and change to 3840x2160 24hz.

And that's what I did here to have custom timing and 12bits.
Okay, I'm not sure you'll ever read this but THANK YOU.

If you own an nvidia GPU and if you are trying to create a custom resolution with custom timings in madvr, this is the way to go, I couldn't to it with another method.

Hope this reply will bump your answer so guys with the same problem as me won't spend hours on Google like I did...

[XMonarchY]

Will there ever be a 12bit fix for custom modes? MadVR custom refresh rate mode tool simply never creates modes above 8bit, even on 12bit displays with 12bit already set in NVidia CP for default refresh rates

[nevcairiel]

I kept telling people for a long time to use CRU to create custom resolutions. It works around any quirks in any APIs, because it just doesn't use any. Let madVR figure out the values if you want, and just put them into the EDID with CRU.

[White Eagle]

is there a step by step guide to doing this with CRU?

[madjock]

is there a step by step guide to doing this with CRU?

Heres one

http://madvr.com/crt/CustomResTutorial.html
https://forum.doom9.org/showthread.php?t=173571

Not used it myself, but I had more success with nVidias, I also do not understand why people want the 10 and 12 bits, the more I read the more I understand it is a waste of time and you are just as well with 8 bit RGB.

[Warner306]

That doesn't mention CRU. It isn't that complicated. You only need to create one custom resolution and simply copy the values calculated by madVR into CRU.

[Charky]

is there a step by step guide to doing this with CRU?

Well the post I quoted 5 posts above explains it quite well.

But TBH it does not work as flawlessly as I hoped.

Every time I reboot I still need to manually pick a 12bit mode (e.g. 2160p30 @ 12 bits) in nvidia CP first. If I don't, the 2160p23 custom res madvr autoswitches to sticks with 8 bits.

However once i've done this first, then when playing a 23p source, madvr will autoswitch from whatever resolution is active (e.g. 2160p60 8 bits) to my custom 2160p23 res in 12 bits.

So in the end it works, but not without bumps, which is to be expected with such a quirky method, I guess...

[madjock]

Well the post I quoted 5 posts above explains it quite well.

But TBH it does not work as flawlessly as I hoped.

Every time I reboot I still need to manually pick a 12bit mode (e.g. 2160p30 @ 12 bits) in nvidia CP first. If I don't, the 2160p23 custom res madvr autoswitches to sticks with 8 bits.

However once i've done this first, then when playing a 23p source, madvr will autoswitch from whatever resolution is active (e.g. 2160p60 8 bits) to my custom 2160p23 res in 12 bits.

So in the end it works, but not without bumps, which is to be expected with such a quirky method, I guess...

But what are you actually achieving here ? apart from seeing a 12 on an OSD ?

[Charky]

But what are you actually achieving here ? apart from seeing a 12 on an OSD ?
If you want to say that 10/12 bits is useless, just say it (and explain why). Being snarky doesn't make you look smart ;)

[madjock]

If you want to say that 10/12 bits is useless, just say it (and explain why). Being snarky doesn't make you look smart ;)

It was a question to be honest, I thought the way you were chasing it was me asking why.

Have you any sources in 12 bit ?

I have read on this forum and many others whilst chasing a supposed better dream by increasing bits and in reality it does nothing and indeed can cause issues.

Search 8bit vs 10 bit in Google and the majority say its not worth it, you cannot get 10bit as nVidia has disabled it unless its for games, you will never see more colours and madVR will never output more than 10bit anyway.

So no not snarky, just thought there was something I was missing, and wondered why people chased the extra hastle of changing settings all the time.

[brazen1]

I don't think there is even a 12 bit panel made for consumers at least nothing mainstream.

[ryrynz]

I switch to 12 bit. After turning off dithering for testing and comparing I could see better graduation in the greyscale ramp, it's minor but it's a free improvement so I'll take it, just test for yourself how your screen handles it.

[huhn]

excuses me? disabling dithering if you only saw minor difference then you should maybe look for a better test pattern.

i even have a troubles imaging an processing error in TV that would be worse then sending 8 bit without dithering.

[nevcairiel]

Search 8bit vs 10 bit in Google and the majority say its not worth it, you cannot get 10bit as nVidia has disabled it unless its for games you will never see more colours

This is just wrong. You can get 10-bit/12-bit output on any modern GPU, depending on monitor support.

madVR will never output more than 10bit anyway.

While this is true for now, its also irrelevant. We use 12-bit because thats what NVIDIA gives you with many HDMI TVs. Its either 8 or 12, and if you can output 10-bit with madVR, then clearly 12 is the only mode to actually preserve those 10-bits, because 8 would reduce it.

[Asmodian]

I switch to 12 bit. After turning off dithering for testing and comparing I could see better graduation in the greyscale ramp, it's minor but it's a free improvement so I'll take it, just test for yourself how your screen handles it.

This is a bad methodology. Leave on dithering when doing all comparisons, you would never watch with dithering off so don't turn it off when testing.

Dithering is often misunderstood, it is not hiding errors, it is how to do the math correctly when viewing/listening to digitally sampled sources.

You can use test patterns with dithering off when trying to understand what your display's internal processing is doing but do not make extrapolations from those tests like you did here. You assume there is a "free improvement" from a test that was not representative of the image you would actually see when using 8 or 12 bit. How does that tell you anything? I get why we all do these overly simply tests but we need to be very careful when interpreting, or disseminating, the results. There can be a lot of misinformation spread based on inappropriate testing methodology.

[edigee]

My Samsung MU7072 is 8bit+frc. NVCP recognizes it as 8 bit only. Full RGB.
Ia that normal?

[huhn]

HDMI can't do more at 60 hz the bandwidth is simply capped.

try 23-30 hz and 12 bit should be an option for full range RGB.

[ryrynz]

This is a bad methodology. Leave on dithering when doing all comparisons, you would never watch with dithering off so don't turn it off when testing.


Of course I did this first, however I couldn't determine a clear result by doing that. So testing this way helped me to determine if this it made any difference at all and because it did i can only conclude if it's not harming quality then it's at least the same or better.. That's good enough for me.