MadVR in use with LG OLED Thread

[chros]

I wouldn't say no one's interested in that here.
Don't get me wrong, but under "interested" I meant those ones who are willing to experiment with their device and not just talk about it :)

[SamuriHL]

Don't get me wrong, but under "interested" I meant those ones who are willing to experiment with their device and not just talk about it :)

I'm still learning the ins and outs of calibration right now so I don't think I'd be a whole lot of help yet. If you figure something concrete with reasonably good steps on how to do it then I can try. I'm still fighting with the HDFury to get Dolby Vision calibrated unfortunately. I need to get back to that soon.

[biship]

Newbie here. Recommended basic MadVr setup are per 1st post?

MadVR:
PC levels (0-255)
display bitdepth: auto
display modes: 2160p23 for movies (Add 2160p60 for regular shows??)
reduce banding artifacts low (default strength), low (fade in/out)
chroma upscaling: NGU AA, medium
image upscaling: NGU Sharp, high

No need to set Display Modes?
Can I leave MadVR HDR settings at 'let madvr decide' or is something recommended here?
Any other MadVR settings I should consider?

My PC is struggling to do any upscaling as it is. (GTX 1050)

[VBB]

Assuming you have an OLED, you should be able to do about the same that I can do with my GTX 960:

PC Levels 0-255
display bit depth 8-bit
display modes set to "do not auto-switch" (we want to stay at 60hz at all times)
reduce banding artifacts low/high
chroma upscaling NGU AA low (no need to go any higher, since we're not going to use 4:4:4 anyway)
image upscaling NGU Sharp high (or lower, depending on what you're upscaling)

You can leave all upscaling at "let madVR decide". Enable "smooth motion=always". HDR set to "pass-through". Make sure DirectX 11 is also checked in both madVR and LAV filters. Set desktop resolution in Nvidia control panel to 2160@60Hz 8-bit RGB Full, power in 3D settings to "adaptive". TV black level to HIGH, HDMI to anything other than "PC". Turn off both TruMotion and Real Cinema.

That should get you some decent results with your card. My GTX 960 rarely goes into the 30s with any content.

[Asmodian]

Great advice for HDR, I mostly agree with your settings but it is depressing that we do not get 4:4:4 with HDR. :(

Also it will look the same on the TV if you use limited YCbCr 4:4:4 output and set the TV to low blacks instead of sending RGB with the TV on high blacks. I think the level of processing going on in the display makes sending full range RGB not helpful in anyway and it may be even worse than YCbCr. These TVs do not do a super high quality RGB to YCbCr conversion internally. This allows me to only change the HDMI label when switching between optimal HDR and SDR modes on the TV.
Edit: I mean setting the desktop resolution in Nvidia control panel to 2160@60Hz 8 bit YCbCr 444 limited, not changing anything in madVR.

HDMI to anything other than "PC".

Game is similar to PC, you need to use something else for HDR content to looks its best.

[mark0077]

Asmodian, I am using your recommended settings for my LG C9 of 2160@60hz 8bit YCbCr 444 limited for standard content and its by far the best with the least amount of banding. (I'm in PC + Game mode).

For HDR what do you recommend now based on your recent experiences? The same mode of 2160@60hz 8bit YCbCr 444 limited in HDR mode seems to make the screen go almost pure green (not sure if its an Nvidia driver bug, it happens only if I don't first set Windows to HD Colour mode so I guess its madVR -> Nvidia driver switch that causes it).

I assume some higher bit depth is required anyways to make best use of the 10bit HDR content. Currently testing with The Lion King UHD Blu-Ray. Seems like I have to go back to RGB or YCbCr 422/420 of varying bit depths? Is there a known best mode on these TVs for HDR to have the least amount of banding, similar to the advice of YCbCr 444 8bit for standard content?

[VBB]

Yes, it's RGB Full 8-bit :)

[biship]

Thanks VBB. Yes I have a OLED65C9PUA, which is connected to my Denon AVR-X3500H. My Windows 10 PC (i7-6850K & GTX 1050, with latest LAVF, MADVR & Nvidia drivers) also connects to the Denon.

On the Denon I've turn on Enhanced Black, Video Mode to Movie, Video Conversion on, but IP Scaler off.
In the 11111 settings on the TV it shows HDCP 1.4 & 2.2 as FALSE. No idea why. It does show 3840x2160P @ 59.92Hz RGB 8BIT BT2020_RGBORYCBCR - which I assumes means HDR10?
Does all this look correct?

Even with minimal settings in MadVR, I get 60ms+ rendering, which results in dropped frames, unless I disable banding processing, dithering, and set all scaling to DXVA2.
Why is MadVR even doing anything to 2160p HDR10 sources? 70ms just for post processing is crazy.
Images:
https://ibb.co/HC7tdfX
https://ibb.co/NCqc63s

[Asmodian]

The same mode of 2160@60hz 8bit YCbCr 444 limited in HDR mode seems to make the screen go almost pure green (not sure if its an Nvidia driver bug, it happens only if I don't first set Windows to HD Colour mode so I guess its madVR -> Nvidia driver switch that causes it).

8 bit RGB full does seem to have slightly less banding than YCbCr 444 limited in HDR10. Another setting to change when switching between SDR and HDR. :(
I do not get any odd issues with the screen going green or anything like that (Win10 1909, 441.66).

I never noticed it before because the banding difference is much less significant than the difference between 8 and 10 bit in SDR. I thought they looked the same but switching back and forth with a color banding test does convince me 8 bit RGB has less banding, still more visible banding than with 8 bit YCbCr 444 SDR but perhaps that is simply due to the increased brightness.

I assume some higher bit depth is required anyways to make best use of the 10bit HDR content. Currently testing with The Lion King UHD Blu-Ray. Seems like I have to go back to RGB or YCbCr 422/420 of varying bit depths? Is there a known best mode on these TVs for HDR to have the least amount of banding, similar to the advice of YCbCr 444 8bit for standard content?

You should not send 10 bit to the display. I never feed my LG C7 or C9 >8 bit. madVR likes the 10 bit input, and keeps much of the extra information with its excellent dithering to 8 bit, but LG does not do a good job handling 10 bit input in any test I have done. At best 10 bit is the same as 8 bit while at worst it has significantly worse banding.

Yes, it's RGB Full 8-bit :)

Do you agree that there is less banding with 8 bit YCbCr 444 limited in SDR even while there is less banding with 8 bit RGB full in HDR? This seems true for both PC and Home Theatre HDR modes. I wish LG was better at image processing, or at least consistent. :p

[SamuriHL]

Interesting. I've been calibrating the C8 using RGB full 8 bit and switching to YcbCr 444 limited for both SDR and HDR. I guess I should try these banding tests you're referring to. Got a link handy for that? I'm watching It Chapter Two right now in HDR YCbCr 444 limited and it looks good to me. RGB looks different. As mentioned before, it's hard to say for sure which is "correct" but I don't get the feeling this display likes RGB all that much. It seems built for consumer device input i.e. YCbCr.

[Asmodian]

It is a pretty subtle difference in banding but I am pretty sure it is real. :o
gradient-perceptual-colored-v2.1 24fps.mkv (http://www.mediafire.com/file/4dia3lixs0n9xyo/gradient-perceptual-colored-v2.1_24fps.mkv/file)
gradient-perceptual-v2.1 24fps.mkv (http://www.mediafire.com/file/z1f4ahg3784lfwz/gradient-perceptual-v2.1_24fps.mkv/file)
Even the bad banding of 10 bit full range RGB is subtle in most real content, these files expose display banding very aggressively.

Edit: Hmm, I am less sure this is actually a useful test in HDR, since it is a perceptual gradient SDR test. Maybe the difference is an interaction with Windows HDR? That seems likely since I saw the same banding in PC and Home Theater mode.
@VBB What did you use to decide RGB was better than YCbCr 444, if you did?

[QBhd]

I used "The Long Night", season 8 episode 3 of Game of Thrones 4k HDR... I think it's about the 16 min mark... Dani and Jon have just started burning up the dead and the banding is un-watchable. I threw every conceivable combination of color space, bitrate, and HDMI mode at this episode. Finally settled on 8-bit RGB and normal HDMI mode.

QB

[SamuriHL]

I saw some pretty bad banding in it chapter two last night, as well. I'll need to compare it against rgb. Come on lg...

Sent from my SM-G975U using Tapatalk

[iSeries]

From what I see, madVR limited 8bit, gpu rgb full 8bit, and TV black level low is the best configuration for an HTPC with an LG TV. A complete 'full' chain has slightly more banding and slightly increased black crush. And as most people will agree, greater than 8bit also results in increased banding, and definitely PC mode results in greatly increased banding.

[mark0077]

Edit: Hmm, I am less sure this is actually a useful test in HDR, since it is a perceptual gradient SDR test. Maybe the difference is an interaction with Windows HDR? That seems likely since I saw the same banding in PC and Home Theater mode.
@VBB What did you use to decide RGB was better than YCbCr 444, if you did?

With those perceptual tests, how are you performing the test? I have to manually set windows to HDR as madVR doesn't trigger it for these tests (I assume because they are not BT2020?). I find when I'm in a smaller window the display stays in HDR mode, as soon as I double click and go fullscreen at all, the TV switches back to non HDR mode.

[Asmodian]

Yep, manually trigger HDR in Windows and keep the player windowed. I think I will stick to YCbCr 444 limited for HDR too though, at least until I get better evidence that RGB full is actually better for HDR. :o

From what I see, madVR limited 8bit, gpu rgb full 8bit, and TV black level low is the best configuration for an HTPC with an LG TV. A complete 'full' chain has slightly more banding and slightly increased black crush. And as most people will agree, greater than 8bit also results in increased banding, and definitely PC mode results in greatly increased banding.

PC mode does not have higher banding if I set madVR to full and the GPU to YCbCr 444 limited. :)

[iSeries]

^ pretty sure I read the TV will do all it's processing at 8bit in PC mode, regardless of rgb/ycbcr input?

[Asmodian]

In my tests the C9 does very well with 8 bit YCbCr 444 limited input in PC mode. Virtually no banding, the same as the non-PC modes. With any other input type (i.e. RGB limited or full) PC mode does have obviously worse banding. I believe the key is to avoid the processing as much as possible when in PC mode.

Given how my C9 handles 10 bit input in PC mode I would be surprised if it was not processing it in 8 bit, it looks exactly like a 10 bit -> 8 bit conversion with no dithering. Chroma subsampling is completely unacceptable for me so I was very happy to find a setting that was virtually banding free in PC mode.

[VBB]

So, after about five hours of testing today, I've found my new favorite combo for both SDR and HDR (you're all gonna laugh):

GPU set to YCbCr 4:2:2, 10-bit, madVR set to PC Levels 0-255, 10-bit or higher, TV set to Black Level Low, non-PC mode

I came to this conclusion after extensive back-and-forth testing for banding. I tried every possibly combination with mainly these two patterns, courtesy of Tyler Pruitt and Stacey Spears, respectively:

https://drive.google.com/file/d/0B68jIlCvW85gWFp0NVUwTFdTNFE/view?usp=sharing
https://drive.google.com/file/d/0B68jIlCvW85gMW5aX21CbU1IOGs/view?usp=sharing

For my testing, I disabled "reduce banding artifacts" and any dithering in madVR, to get a fairly unprocessed image. I agree with Asmodian that YCbCr 4:4:4, 8-bit is the best option for PC mode (both SDR and HDR), with RGB Limited, 8-bit a close second. Unfortunately, there is still quite a bit of banding in PC mode compared to non-PC mode. The added benefit of 4:4:4 is just not worth it to me after seeing how much smoother YCbCr 4:2:2, 10-bit is in non-PC mode. I had never even tried this combo until today, not realizing that 10-bit was in fact an option with Nvidia, but only with this particular combo. Now, as most of you know by now, there really isn't any benefit to using 10-bit over 8-bit with dithering, but this gave me the smoothest gradients of all the combos I tested. There is virtually no banding even without dithering.

Coming from almost two years of RGB Full, 8-bit usage, I had to check that all my other settings were still OK. Used an all-black 0% pattern with my nose less than an inch from the screen in total darkness. I found that I had to decrease Brightness on my C7 by 3, from 51 down to 48, which I thought was odd. In HDR mode, I had to decrease it by just 1, from 51 down to 50. I should probably also redo my LUTs. A great pattern to check black level is this one by Wizziwig:

https://www.avsforum.com/forum/attachment.php?attachmentid=1661825&d=1473909205

If anyone else is interested in testing all this, I'm curious what you'll find, especially with the newer LGs. Don't forget to turn dithering back on in madVR when you're done. I'ma go watch some stuff now. Will report back if I notice anything weird.

[SamuriHL]

I watched the long night with pc mode YCbCr 444 limited earlier and there was wicked banding in most of the fire scenes. I really really like 444 but the banding it's quite horrific with my c8 for HDR. I didn't notice it as much for sdr and the benefit to sdr upscaling is huge. Get it together lg! Ugh

Sent from my SM-G975U using Tapatalk

[QBhd]

The Long Night 4k HDR... it will hammer your preferred banding setup.

QB

[VBB]

I watched the same thing earlier, in both 1080p SDR and 4K HDR. It was smooth.

[SamuriHL]

Yeah I figured it was going to be a good test. Lol. But I also see banding in the spears and munsil uhd demo. I'll do some more playing later and see if I can eliminate it. It would be fantastic to get it to a watchable state at 444.

Sent from my SM-G975U using Tapatalk

[chros]

It would be fantastic to get it to a watchable state at 444.
In PC mode? :D (Just kidding.)

PC mode does not have higher banding if I set madVR to full and the GPU to YCbCr 444 limited. :)
But we know that's not even proper 4:4:4 (although really close to it), do we? :)

It's not just banding, but posterization in near black scenes in HDR (e.g. <20 nits), it's just absolutely rubbish :D (Read from here (https://forum.doom9.org/showthread.php?p=1883108#post1883108).)
Note, that I still use PC mode with RGB full 12 bit (I gave up on experimenting with it, all of them had some issues), and just try to enjoy it :)

[QBhd]

You say that now, but something like The Long Night is truly un-watchable like that :)

QB

[Asmodian]

But we know that's not even proper 4:4:4 (although really close to it), do we? :)

What does really close to but not proper 4:4:4 mean? There are odd pixel interaction artifacts, something like crosstalk, but it is not an issue with chroma subsampling.

It's not just banding, but posterization in near black scenes in HDR (e.g. <20 nits)

For HDR I gave up on 444 a long time ago. It is more than just posterization and banding, the color gamut is all messed up too. I use Home Theater for HDR content. :(

[mark0077]

It seems like most are coming to the conclusion so that YCbCr444 8bit limited shows least banding for SDR and HDR. My fear is we are not ever seeing true 10bit content from these UHD movies, only an 8 bit dithered version of it (albeit very good through madVR). Is there some calibration accuracy test that can compare this to non PC mode with the higher bit depths to see if they show a true 10 bit image with the full dynamic range that brings? Still feels wrong to have to "settle" for 8 bit when the original content is 10 bit doesn't it?

[SamuriHL]

You say that now, but something like The Long Night is truly un-watchable like that :)

QBYou ain't kidding!

Sent from my SM-G975U using Tapatalk

[SamuriHL]

I genuinely don't worry about bit depth. In a blind comparison you'll be hard pressed to tell the difference between properly dithered 8 bit vs native 10 bit. The problem is that these panels don't seem to handle higher bit depth properly. I'd not get hung up on that issue.

Sent from my SM-G975U using Tapatalk

[chros]

Is there some calibration accuracy test that can compare this to non PC mode with the higher bit depths to see if they show a true 10 bit image with the full dynamic range that brings?
There are: in Mehanik's HDR test patterns from avsforum, there's a section for this. Take a look.

[chros]

You say that now, but something like The Long Night is truly un-watchable like that :)
:D I watched it, so it's watchable, but you can throw almost any HDR content to it and it behaves the same way.

What does really close to but not proper 4:4:4 mean? There are odd pixel interaction artifacts, something like crosstalk, but it is not an issue with chroma subsampling.
I'm not sure that I understand what you mean, but does it matter in the end if there are pixels where they shouldn't be? :)
(And I didn't know you only talk about SDR.)

[VBB]

No comments on what I posted last night? Sure, you'd have to backtrack a page, but come on! ;)

[iSeries]

@VBB how do you find ycbcr 422 10bit vs limited>full RGB 8bit>black level low? I quickly looked at this on my c8 and don't really see any difference in gradation between them

[VBB]

I found it to be the smoothest. Trust me, I would have never switched from RGB to YCbCr if I hadn't seen this with my own eyes. How did you test? PC mode only, or both? Unsurprisingly, this combo is not smooth at all in PC mode.

[SamuriHL]

I haven't had a chance to test it yet. I am curious, myself.

[SamuriHL]

So, after about five hours of testing today, I've found my new favorite combo for both SDR and HDR (you're all gonna laugh):

GPU set to YCbCr 4:2:2, 10-bit, madVR set to PC Levels 0-255, 10-bit or higher, TV set to Black Level Low, non-PC mode


Yea, this does look quite good but you should be aware of something. "10 bit" is being sent as 12 bit by the GPU regardless of what it's telling you. I'm testing now with more GoT and it looks very good so it might become my new "goto" settings, too, but, more testing is needed. Like a lot more. Nonetheless, this has quite a bit of potential.

[iSeries]

^ does using the hidden option to output ycbcr from madVR improve things or make things worse?

[SamuriHL]

I'm not sure I know how to do that. I can say that the Envy is going to output YCbCr by default so it might actually be worth trying since the Envy code was merge. Care to share how to do that?

[iSeries]

I think it's something like create an empty folder or a file called ycbcr

[SamuriHL]

Interesting. We may have to mess with that. I really, truly feel the LG OLEDs do NOT like RGB input. So that might be useful. I don't have time to play with that tonight.

[nevcairiel]

That option would be rather rarely useful as you need to convince your TV thats its receiving YCbCr when everything is telling it that its receiving RGB (which very few display can even do), plus it screws up anything but madVR output, ie. Windows and player UI to boot.

[SamuriHL]

I'm confused with what you're saying. Are you saying using the ycbcr option in madvr and setting the GPU to RGB (why would anyone do that?) would cause that? Or are you saying just sending YCbCr from the GPU causes it? From what I can see, the GPU is sending YCbCr and the TV is accepting that. I think iSeries was proposing that instead of madvr outputting RGB and having the GPU convert it before sending it to the TV, that we'd have madvr output YCbCr instead. Would that not work? Is that what you're saying?

[VBB]

He's talking about this here:

Hidden Options
An empty folder or file in the madVR program directory with a specific name will trigger a hidden option, these are options you probably do not want to use.

YCbCr: Outputs YCbCr data instead of RGB, as if it was RGB.

I have not messed with that, so I can't say if it's beneficial.

[nevcairiel]

I think iSeries was proposing that instead of madvr outputting RGB and having the GPU convert it before sending it to the TV, that we'd have madvr output YCbCr instead. Would that not work? Is that what you're saying?

That doesn't work because the backbuffer (ie. the screen) is always in RGB. So when madVR outputs YCbCr to the backbuffer, then the GPU still thinks that its RGB, and if the GPU tries to convert it, it would screw it up - so it has to treat it like RGB.

Hence, everything thinks its handling RGB, the GPU and the screen, so you have to convince one of them that its not RGB, and thats not easily possible.

Basically, a PC is not designed for a direct YCbCr path.

[QBhd]

Basically, a PC is not designed for a direct YCbCr path.

That's the main reason why I settled on 8-bit RGB with the default HDMI label on the TV.

QB

[Asmodian]

I'm not sure that I understand what you mean, but does it matter in the end if there are pixels where they shouldn't be? :)

Chroma subsampling is blurring the image across the entire screen while some color fringing on the edge of a solid yellow regions or similar is a very different issue. Changing the subsampling done by the TV does not change the artifacts, they are unrelated issues, so it seems wrong to bring them up as if they were related. :p

Is there some calibration accuracy test that can compare this to non PC mode with the higher bit depths to see if they show a true 10 bit image with the full dynamic range that brings? Still feels wrong to have to "settle" for 8 bit when the original content is 10 bit doesn't it?

10 bit does not increase the dynamic range in anyway. I did do accuracy tests with Calman Video Pro and 10 bit does not increase accuracy at all. All it does is decrease the magnitude of the dithering noise but humans tend to perceive subtle noise as sharpness. The very slightly higher noise floor of dithered 8 bit compared to dithered 10 bit is not a significant negative, if noticeable at all. Certainly not on these LG OLEDs.

The video processing path will be different for different bit depths so various displays will handle 10 bit better or worse but the total dynamic range should never change.

[SamuriHL]

That doesn't work because the backbuffer (ie. the screen) is always in RGB. So when madVR outputs YCbCr to the backbuffer, then the GPU still thinks that its RGB, and if the GPU tries to convert it, it would screw it up - so it has to treat it like RGB.



Hence, everything thinks its handling RGB, the GPU and the screen, so you have to convince one of them that its not RGB, and thats not easily possible.



Basically, a PC is not designed for a direct YCbCr path.That sadly makes sense.

Sent from my SM-G975U using Tapatalk

[chros]

Chroma subsampling is blurring the image across the entire screen while some color fringing on the edge of a solid yellow regions or similar is a very different issue. Changing the subsampling done by the TV does not change the artifacts, they are unrelated issues, so it seems wrong to bring them up as if they were related. :p
:) I understand what you mean now, but is the end result proper 4:4:4 or not? :D Believe me, I'd be one of the happiest chap if it was :)

No comments on what I posted last night? Sure, you'd have to backtrack a page, but come on! ;)
I read it and I'm happy for you if you are happy :) Were you talking about SDR only and not PC mode, right?
I'm like Asmodian (probably worse, with HDR), I won't abandon PC mode :)
(We are busy with playing calibration, really frustrating :) )

[SamuriHL]

PC mode is a mess. I wanted very badly to get it to work but it's not usable imo. If you can live with the issues that's awesome. For me even with my ub820 it's a mess.

Sent from my SM-G975U using Tapatalk

[VBB]

I read it and I'm happy for you if you are happy :) Were you talking about SDR only and not PC mode, right?
I'm like Asmodian (probably worse, with HDR), I won't abandon PC mode :)
(We are busy with playing calibration, really frustrating :) )

I was talking about both SDR and HDR, non-PC mode. There is no way to get PC mode to be smooth. So, yeah, I'm happy, for now anyway.

I used to spend days with the meter in front of my screens, but with the C7 I've only done it a couple of times. My SDR LUT is still good, although it might be time to redo it. As far as HDR is concerned, I tried when my set was new, but I realized quickly that a LUT via DisplayCAL just wasn't possible due to the instability of the display. So I dialed in the white point and called it a day ;)

I'll start playing with calibration again once I get a better meter and the C10 next year.