MadVR in use with LG OLED Thread

[j82k]

Originally Posted by janos666

I see a little bit of magenta but nothing crazy and I can't capture it with a phone (55C8, 05.10.40 EU, Cinema mode, not calibrated, PC source 12bit Full RGB).

That looks better than on my C8.
Now I'm wondering if this is a panel variance thing. My C8 is way too dark in the low end and the only way to solve this for HDR without losing true black is to increase 2P Low RGB. Maybe this has something to do with it.

In my case it's definitely not a PC problem as these scenes look similar played from USB or another playback device.

So you're on 05.10.40. I assume that's an engineering FW version?

[j82k]

Originally Posted by chros

Now I understand what you meant, and I replicated the same thing on B8: SDR converted image looks much more natural (using ISF Bright room @140 nits in Wide gamut mode gamma 2.2, plus external 3dlut). And it's not firmware related, since I'm still on 4.10.25.
Nice find, again!

I always wondered whether we need the BT.2020 mode (small logo at the top on the info screen) using HDR, since we can reach close to DCI-P3 in Wide SDR mode as well.

And you know what? Remember when I told you that PC mode is fine with bright images but it's getting ugly with dark images? Probably it's the same problem, but exaggerated by PC mode.

So, the question is: is there any way to trigger HDR mode without enabling BT.2020 mode? This could solve our issue using madvr "output HDR" without introducing new ones

What I've been doing the past weeks was to tone map to SDR, madVR set to DCI-P3, TV to wide. Also ModuleHDR in the service menu turned on which gives me like 550 nits SDR peak brightness with Oled/Contrast 100. Not sure how correct this is but I prefered that to HDR output even though it's 200 nits less.

[chros]

Originally Posted by j82k

Now I'm wondering if this is a panel variance thing.

I'm pretty sure it has nothing to do with panel variance: mine is a normal panel (the original one, meaning it hasn't been replaced), and I experience the same thing with B8. (And I know that @janos666 thinks that B8 is way worse in HDR than C8 )

Originally Posted by j82k

So you're on 05.10.40. I assume that's an engineering FW version?

Yes, it is, he is a brave man!

Originally Posted by j82k

What I've been doing the past weeks was to tone map to SDR, madVR set to DCI-P3, TV to wide. Also ModuleHDR in the service menu turned on which gives me like 550 nits SDR peak brightness with Oled/Contrast 100. Not sure how correct this is but I prefered that to HDR output even though it's 200 nits less.

Interesting! I have never tried this. @janos666 can tell more about its flaws.

Originally Posted by janos666

I see a little bit of magenta but nothing crazy and I can't capture it with a phone (55C8, 05.10.40 EU, Cinema mode, not calibrated, PC source 12bit Full RGB).

Have you compared to SDR mode as well? Switch between e.g. passthrough and HDR2SDR pixelshader (not "output HDR"), it should be clearly visible. Pick a scene around 10 nits

Originally Posted by janos666

There is but the color gamut is limited to Rec709.

I don't think so, only if Auto is selected for the gamut. But you can select Wide in Normal mode, and we can use Standard preset to get Wide gamut in PC mode as well. This should give us the native panel gamut, just as in SDR mode.

Originally Posted by janos666

I have a bug with my AMD Ryzen+Vega notebook where Rec2020 fails to trigger in certain resolutions and it's immediately visible (colors are way under-saturated) but HDR still works (the full dynamic range is there).

Wow! That's really good news, thanks!

Originally Posted by janos666

Yes because these are always Rec2020 colors. The P3 metadata merely informs you (and the display, in case it's interested) that the mastering display had P3 primaries (but the content can still have colors outside the P3 gamut). So, no, Rec709 metadata + Wide probably won't work (well, I never tried but it shouldn't, unless it works in some unexpected way).

I know what you mean, but I think it can work: since madvr does the gamut mapping (desaturate dropdown) even when "output HDR" is selected, then we don't need the broken LG BT.2020 gamut processing, instead we can set the display to its native gamut (Wide option).

I just asked madshi to add a switch for us. If you like the idea then drop there a comment, he'll more likely implement it if more people ask for it

[janos666]

Originally Posted by SamuriHL

I've been using 760 without ill effect so far. I can calibrate it for 700 and see what that looks like in comparison. Figures LG gimped HDR10 support on this thing. LOL

It's not about clipping, banding or similar artifacts but accurate TRC tracking. The TV's processor does the tone-mapping with 700 nit peak in mind. If your panel is capable of and set to 800 nit then the whole curve has a positive offset (every color, even dark tones are more or less brighter than reference). madVR's mapping (when set to your real peak white but above 700 nit) won't correct that offset.
Another reason to set 700 nit in madVR is that the TV thinks the panel is 700 nit and thus this is the point where it should start doing hard-clip instead of roll-off and madVR theoretically sends this number as MDL=maxCLL (so your TV should do hard-clip at it's nominal peak with 700). Anything less and you waste the dynamic range of the hardware, more and your TV start doing roll-off.

Originally Posted by SamuriHL

Oh goodie. I'm going to have a blast huh? Well, I wanted one for a few reasons so I will spend the time to get it connected and working. I wasn't planning on letting Calman set the metadata for DV anyway as I wasn't sure if they added support for the Integral 2 yet or not. Last article I saw on it said support was coming. You wouldn't happen to know the proper metadata offhand would you? This is what I found:

§ Paste (but don’t do anything for now) the follow HEX code into the CUSTOM HDR IF: field: 81:01:18:36:03:0c:00:20:01:00:00:00:00:00:00:00:00 :00:00:00:00:00:00:00:00:00:00:00:00:00
§ Confirm the following code is in the CUSTOM AVI IF: field: 00:08:00:00:00:00:00:00:00:00:00:00:00

Well, actually, I have the original Integral (rev2) but I assume the HEX codes are universal. I forgot the Integral 2 has a new CP GUI, so probably needs specific CalMAN updates as well.
CalMAN sets what you quoted. I had to use these (only the AVI IF is different).

Originally Posted by j82k

What I've been doing the past weeks was to tone map to SDR, madVR set to DCI-P3, TV to wide.

Yeah, that should work as far as SDR Wide is practically (almost) DCI-P3.

Originally Posted by j82k

Also ModuleHDR in the service menu turned on which gives me like 550 nits SDR peak brightness with Oled/Contrast 100. Not sure how correct this is but I prefered that to HDR output even though it's 200 nits less.

Last time I checked (may be that wasn't on a C8 but a B8 or C7) ModuleHDR=ON will mess up the color accuracy of the SDR mode as the W starts to over-shine the RGB set and you get that increasingly less saturated "dome" at the top of the dynamic range (where the W can go higher but the R,G,B are already maxed out, so the panel can get brighter but won't deliver the correct color). This is different with Alpha7 and Alpha9 branded TVs. The A7 starts to de-saturate the colors at lower brightness (although I think this has nothing to with the processor but done by the panel driver where the RGB->WRGB magic happens, yet the Alpha branding still tells you which mode is utilized in HDR on the given model).
Thus, wouldn't set it to always ON (unless you really know what you are doing but CMS softwares don't like these non-additive color spaces, so it's hard, if not practically impossible, to create a 3DLUT correction for this). Always Off could be more useful than always On (but only on 2019 models or when yo want a 3DLUT correction).
I don't know why SDR 100/100 doesn't reach the HDR peak brightness with this ModuleHDR switch turned ON, I guess the OLED Light slider values are mapped to different lower level values in SDR (and SDR is clipped below actual 100%). Or may be it's clipped by some ABL (even on tiny windows) which behaves differently in SDR.

Originally Posted by j82k

So you're on 05.10.40. I assume that's an engineering FW version?

Yes but I am sure it has nothing to do with this. I bet this issue spans through several model years, SoC revisions and firmware versions. It was always clearly visible in PC mode, thus it's probably always there for a varying degree.

Originally Posted by j82k

Now I'm wondering if this is a panel variance thing. My C8 is way too dark in the low end and the only way to solve this for HDR without losing true black is to increase 2P Low RGB. Maybe this has something to do with it.

May be your panel has slightly more near-black banding and/or white balance fluctuations even with any and all kind of SoC processing turned off (e.g. put into calibration mode and fully reset). And any kind of SoC processing (except dithering or de-banding) only reduces the (perceived) precision, especially the user menu controls. It would be interesting if you looked at these scenes with the RGB Low/High controls at default (yeah, even the High because the user menu WP setting is not flawless either, you should use the EZ-ADJUST controls if you need to set this). And never touch the CMS in HDR!

[chros]

Originally Posted by j82k

What I've been doing the past weeks was to tone map to SDR, madVR set to DCI-P3, TV to wide. Also ModuleHDR in the service menu turned on which gives me like 550 nits SDR peak brightness with Oled/Contrast 100. Not sure how correct this is but I prefered that to HDR output even though it's 200 nits less.

Originally Posted by janos666

Last time I checked (may be that wasn't on a C8 but a B8 or C7) ModuleHDR=ON will mess up the color accuracy of the SDR mode as the W starts to over-shine the RGB set and you get that increasingly less saturated "dome" at the top of the dynamic range (where the W can go higher but the R,G,B are already maxed out, so the panel can get brighter but won't deliver the correct color). This is different with Alpha7 and Alpha9 branded TVs. The A7 starts to de-saturate the colors at lower brightness ... .
Thus, wouldn't set it to always ON (unless you really know what you are doing but CMS softwares don't like these non-additive color spaces, so it's hard, if not practically impossible, to create a 3DLUT correction for this). Always Off could be more useful than always On (but only on 2019 models or when yo want a 3DLUT correction).

I just quickly tested it: SDR gamma tracking is screwed up completely, and we haven't even talked about color accuracy yet

And as @janos666 said, I don't think that even a 3dlut can correct that amount of panel instability (although can moderate the issue). (I just quickly enabled my existing SDR gamma 2.2 3dlut as well to see the effect.)
Otherwise it didn't look that bad.

Anyway, I hope madshi will implement that checkbox, so we can try it out.

[janos666]

Originally Posted by chros

Have you compared to SDR mode as well? Switch between e.g. passthrough and HDR2SDR pixelshader (not "output HDR"), it should be clearly visible. Pick a scene around 10 nits

I tried but all colors on these images look significantly different to start with, so it's not that easy to guess the truth. j82k's Panasonic images show some magenta as well, just a lot less...

Originally Posted by chros

I don't think so, only if Auto is selected for the gamut. But you can select Wide in Normal mode, and we can use Standard preset to get Wide gamut in PC mode as well. This should give us the native panel gamut, just as in SDR mode.

The Color Gamut presets work differently in SDR and HDR. In SDR Wide=Native. In HDR Auto is supposed to yield correct colors but Extended and Wide "boost" the saturation (more vibrant colors than reference, if possible within the Native boundaries, clipped otherwise).
If HDR Wide was Native then colors would look under-saturated since the content is Rec2020 (this would be the opposite of watching Rec709 content in SDR Wide where the content gamut is smaller and thus it appears over-saturated --- in this case the content gamut is bigger, so...).
It's difficult to guess what really happens in HDR mode. We have a static 3DLUT in use (which could do the Rec2020->Native mapping) but colors are still limited to Rec709 when we don't send the Rec2020 metadata, so either some math-based processing is involved on top of the 3DLUT or the TV switches to a different 3DUT (currently non-programmable factory one with Rec709->Native mapping). Hmm... Some experimentation could answer this...
We know that colors look under-saturated in HDR with a neutral 3DLUT (if you exit calibration mode after a DDC reset), so I guess the 3DLUT is Rec2020->Native, not P3->Native and thus there is no Rec2020->P3 real-time math. So, in case of no Rec2020 metadata, we either have the Rec2020->Native 3DLUT + Rec709 math or the TV switches to a non-programmable factory Rec709->Native 3DLUT.
Anyways... if you want Native gamut in HDR, you have to reset the 3DLUT via DDC and keep the Rec2020 metadata in the signal. And then convert externally from Rec2020 to P3 (but not to Rec709).

Or may be... just may be... HDR Wide could be Native with no Rec2020 matadata in the signal. (I haven't though about that. Sorry.) But you have to check if that's the case. I somehow doubt this.

[j82k]

Originally Posted by chros

I just quickly tested it: SDR gamma tracking is screwed up completely, and we haven't even talked about color accuracy yet

Interesting, I'm pretty sure when I tested this HDR peak brightness didn't increase with HDR Module On. So the B8 apparently behaves differently.

Also for some reason the sports preset is significantly brighter than all the other presets with HDR Module On,even with color temperature set to W50. Too bad it's such a crappy preset so that's useless.

Originally Posted by janos666

j82k's Panasonic images show some magenta as well, just a lot less...
.

Yes I think this is more of a saturation problem and not related with the pc-mode banding...

[janos666]

I came back with a few more images (with a better camera this time): HDR passthrough vs. madVR SDR conversion

Originally Posted by j82k

Yes I think this is more of a saturation problem and not related with the pc-mode banding...

Banding can cause saturation errors. It means you get the same amount of a color component across a range of input values. That means sometimes you will have way too much or way too little of a component if it's severe enough. The effect is not limited to the gray scale (where it messes up the white balance) but affects random mixed colors as well. And in this case, we also have the W-magic to make it even more difficult.
But it could be something else, of course. May be this happens around the edge of the native gamut where the processor tries to clip to the closest color. But "closest" could mean a lot of things, depending on the color space and logic. It's not necessarily closest perceptually for humans.

[j82k]

Just did a quick comparison, I think on my C8 the difference is more significant.

Passthrough:


HDRtoSDR (madVR set to DCI-P3, TV Wide)


The HDRtoSDR picture looks much better and closer to what it looks like on the Panasonic.

It looks a bit exaggerated on the photo but to remove the oversaturated cyan in passthough mode I would have to lower color to like 30 or so and that of course would mess up brighter scenes. And like I said before, playing the file from the TVs USB looks similar, so not a pc problem...

[VBB]

So guys, I've been following the conversation, and I still don't understand why forcing BT.2020 via madVR is necessary with the OLEDs.

My C7 uses non-PC mode set to WIDE for SDR, AUTO for HDR. I have a 3D LUT for SDR in every slot in madVR (I know that's not necessary). I currently don't use tone-mapping, and HDR in madVR is set to pass-through.

Am I missing something in this scenario by NOT having the "force BT.2020" option checked? If the TV is already set to its widest gamut, what difference would it make?

[huhn]

HDR passthrough automatically triggers a kind of send BT 2020 at the end device because it know the image has a gamut up to bt 2020.

if you tone map then there is a good reason to use BT 2020 and with projector it's kind of a needed function.

[SamuriHL]

More specifically if you tone map without the output HDR option enabled. If you do use it there's no need to enable the send bt.2020 option and in fact you shouldn't if you view any rec.709 content else it gets sent as bt.2020.

Sent from my SM-G975U using Tapatalk

[janos666]

I managed to set up the HDR10 Game mode with a neutral 1DLUT (started with full DDC reset) and a 3DLUT created with DisplayCAL (33 linear gray steps + R,G,B at 50% [safely below the RGB clipping point around 66%], Rec2020 + gamma 2.2 targets and a shaper+matrix profile).
Result: zero magenta artifacts in Lion King (the CalMAN 1D+3D LUT calibrated Technicolor profile has the most of it) and no under-saturation either (the factory Cinema mode has minimal magenta artifacts but it often shows grays instead of blues on those areas).
But Jolie's face is even further from the SDR conversion results (same camera settings but I held it in my hand this time instead of restoring the original position with the tripod and all...):





However, this was my very first attempt with DisplayCAL, so some tweaking of the test chart could improve things.

[SamuriHL]

So far no luck in getting the Integral2 to work with DV. It switches the mode to DV but black screen only and I have to reset the EDID to get it back. There can be a whole myriad of reasons as to why this isn't working unfortunately. SIGH

[SamuriHL]

The ingegral2 is starting to massively piss me off. It's broken my ability to do RGB 8 bit FULL on my HTPC. It turns the screen completely purple. WTF?! Setting it to limited is fine. If I take the Integral2 out of the mix, it's fine. It's making me very very irritable. Not to mention I can't get the goddamn thing to enable DV which is the big reason I bought it. Fix every HDMI issue...guaranteed? Um, RIGHT!!!!!!!! I'm really irritated now.

EDIT: Ok I fixed the RGB 8 bit FULL issue. This device is TOUCHY. It and I are going to have to get acquainted slowly. It's very powerful but good lord. Still don't have DV working properly. I'd like to get that done so I can calibrate it. It needs to be done before Friday.

[janos666]

Originally Posted by SamuriHL

So far no luck in getting the Integral2 to work with DV. It switches the mode to DV but black screen only and I have to reset the EDID to get it back. There can be a whole myriad of reasons as to why this isn't working unfortunately. SIGH

Try setting all hardware dip switches to the middle position (software controlled), then in the GUI software, set HDCP to 1.4 (wasn't necessary for me), Audio to Jack for HDR10 and HDMI for DolbyVision (this is how it worked for me on the old Integral 1), DolbyVision is fine with most EDID settings (fairly high-bw/spec) but HDR10 needs the Default (or similarly low-bw/spec).
Regardless of this injection thing, most of the high-bandwidth, high-spec EDIDs cause issues at high pixel clocks (high res, high refresh rate) with my Mobile Vega 8 (notebook HDMI output) but work fine with my desktop GTX1070.
I ended up using 1080p24 for DV and 2160p30 for HDR10 (couldn't get any other res working with DV and interestingly 1080p24 doesn't work with HDR10 injection, crazy...). Some other resolutions cause black screen or fail to register Rec2020, etc. But most of these issues are related to the AMD GPU. Most things work flawlessly with an nVidia card (even things which shouldn't work according to manual).
Oh, and update the firmware unless you are already at the latest. And try the DV AVI IF I linked a few days ago (keep in mind that these are for given display modes, like 1080p24 RGB and won't work with other formats).

[chros]

Originally Posted by j82k

Just did a quick comparison, I think on my C8 the difference is more significant.
...
The HDRtoSDR picture looks much better and closer to what it looks like on the Panasonic.

It looks a bit exaggerated on the photo but to remove the oversaturated cyan in passthough mode I would have to lower color to like 30 or so and that of course would mess up brighter scenes.

Same here with the B8.

Originally Posted by janos666

I managed to set up the HDR10 Game mode with a neutral 1DLUT (started with full DDC reset) and a 3DLUT created with DisplayCAL (33 linear gray steps + R,G,B at 50% [safely below the RGB clipping point around 66%], Rec2020 + gamma 2.2 targets and a shaper+matrix profile).
Result: zero magenta artifacts in Lion King (the CalMAN 1D+3D LUT calibrated Technicolor profile has the most of it) and no under-saturation either (the factory Cinema mode has minimal magenta artifacts but it often shows grays instead of blues on those areas).
But Jolie's face is even further from the SDR conversion results (same camera settings but I held it in my hand this time instead of restoring the original position with the tripod and all...)

Thanks for testing it!
So do you say that is better now or "just" different? Could you clearly see the issue in person in uncalibrated mode? Honestly, I could barely see the issue on your original screenshots, but here the difference is big (= clearly visible).

I just tried to explain madshi what needs to be done if it makes sense at all.

And here is the result of recent HSTM experience.

[SamuriHL]

Originally Posted by janos666

Try setting all hardware dip switches to the middle position (software controlled), then in the GUI software, set HDCP to 1.4 (wasn't necessary for me), Audio to Jack for HDR10 and HDMI for DolbyVision (this is how it worked for me on the old Integral 1), DolbyVision is fine with most EDID settings (fairly high-bw/spec) but HDR10 needs the Default (or similarly low-bw/spec).
Regardless of this injection thing, most of the high-bandwidth, high-spec EDIDs cause issues at high pixel clocks (high res, high refresh rate) with my Mobile Vega 8 (notebook HDMI output) but work fine with my desktop GTX1070.
I ended up using 1080p24 for DV and 2160p30 for HDR10 (couldn't get any other res working with DV and interestingly 1080p24 doesn't work with HDR10 injection, crazy...). Some other resolutions cause black screen or fail to register Rec2020, etc. But most of these issues are related to the AMD GPU. Most things work flawlessly with an nVidia card (even things which shouldn't work according to manual).
Oh, and update the firmware unless you are already at the latest. And try the DV AVI IF I linked a few days ago (keep in mind that these are for given display modes, like 1080p24 RGB and won't work with other formats).

Yeah I read the link you gave very carefully and set up 1080/24 with no windows scaling (set to 100%). Could not get dv to show up. I didn't try hdcp 1.4 so I'll try that. I factory reset the tv list night so I still need to redo HDR. I got sdr rec709 dialed in very nicely. And yes on the latest firmware as well. Not sure I had all the switches set to the middle. Maybe that's part of the issue too. Finicky device. Doing this on my rtx 2070. Thanks for the help!!

Sent from my SM-G975U using Tapatalk

[janos666]

Originally Posted by chros

Thanks for testing it!
So do you say that is better now or "just" different? Could you clearly see the issue in person in uncalibrated mode? Honestly, I could barely see the issue on your original screenshots, but here the difference is big (= clearly visible).

On the Lion Kind images, I mostly see (in person) a little de-saturation (some blue shades turn into gray) and minimal magenta (I really have to look for it) in factory Cinema and a little bit of meganta error in CalMAN 1D+3D LUT calibrated Technicolor (I don't have to concentrate and stare at it but it's not too bad), zero issues in DisplayCAL 3D LUT (+ not factory but cleared, neutral 1DLUT) calibrated Game mode (absolutely no sign of any mageta, the entire scene is deep blue and there is no grayish de-saturation on the shadows either, it goes from blue to black where there is no bright rock).

I tried to re-fine the DisplayCAL method todays (using a bigger patch set) but realized that CalMAN expects an AutoDesk/Kodak 3DLUT with an actual 16-255 range, so the DisplayCAL LUT either clips the upper end (significantly reduces the peak white) or crushes the blacks (the first one wasn't so obvious at first). This method would require LS DeviceControl...

My guess is that we should see zero magenta during that Lion King hyena scene and it's a human-subjective error in the CMS which creates the 3DLUT (factory one included) or an inaccuracy of the TV's processing (the 1DLUT and/or the user menu WB/gamma controls) similar to the more prominent magenta banding in PC mode.
I remember having similar issues with yCMS (a software which was put together by a member of this forum to create 3DLUTs for madVR before it was discontinued in favor of the more established ArgyllCMS).
When the content has colors very close to or out of the display gamut, it's not obvious what's the human-perceptually closest available color to use instead. Close proximity is a spear segment in a cube and what's equidistant in an YXZ space isn't necessarily so in some other color space like Jab. So, this might looks mathematically closest in a certain space to some algo but subjectively a human might prefers to see yellowish compromise rather than purplish compromise, etc. There is no absolute objective perfect solution for this. I think ArgyllCMS uses Jab or Dab (or something like that) space to make these kinds of clipping decisions. CalMAN might use another working color space (as well as whatever CSM is used for the factory LUT).

And the de-saturation with the factory LUT could actually be intentional. Some calibration guides suggest doing this on purpose to better match the OLED screens to the "standard" LCD colors (where you have no real black and the gray "background noise" de-saturates dark colors). But, of course, this could be a random fault in the factory state, especially after all those "macro block" fixes which altered the near-black dithering and distorted the factory calibrations for some degree.

[SamuriHL]

Ok, I got HDR recalibrated. That was a chore. LOL You are 100% right about NOT going above 700 nits on these panels. I watched it build my curve and saw what it was doing to everything above 700. Um, no. Just no. LOL I set madvr for 700 and there is definitely a difference. However, the stupid factory reset turned my dynamic tone mapping back on which explains why it had some trouble calibrating the mid-range of the greyscale. I will have to do it one more time with that turned off cause I do not like the LG DTM. Nonetheless, the picture quality with neo-xp's latest curve and this latest recalibration is absolutely freaking outstanding.

I'll get to Dolby Vision again tomorrow. Or at least make another attempt. I got some longer USB cables coming which will help me not have to keep rearranging my living room in order to setup a calibration session.