MadVR in use with LG OLED Thread

[chros]

So basically you don't use anything :D I mean, at 700 nits and above there's no work to be done with the majority of contents.
But madvr can do some crazy things as well, I sometimes switch between passthrough and pixeshader and the latter can be darker than the former :)

The 700 nits came from the calibration gurus who stated that the built in curve highest peak is set to 700 with 2018 LGs (just like with 2019 models by default). And on top of this the B8 has 200 nits less than C8 so for me it's 700.

[QBhd]

Again... That's probably a 10% Window... Not the Peak that madVR will spit out... Super small super bright highlights will be capped to 700 well below what the display is capable of. That's why the measured 10% Window is actually brighter than 700 (730ish iirc)... I would guess that is the TV trying to pump up brightness to what it thinks is best.

QB

[SamuriHL]

I actually don't think Dynamic Target Nits is for us. I have tested it on a few videos and it just makes the image dull. The one scene that jumps out was the Netflix Breaking Bad movie at the beginning where they are outside... DTN just made it flat and dim with almost no Dynamic Range. I would guess (not tested yet) that Fire Tweaks and Highlight Recover are still good to use... Sky stuff maybe.. And we don't need Contrast Recovery since we have high peak and perfect Black.

QB

I suspect you may be right. But we definitely want to play around much more.

[QBhd]

I think we should separate two things. chros may be right and that 700 is the internal target. What I suggest we stop doing is equating the curve to the peak. From what I understand, for a projector it is equal. For OLED it is not. Maybe for Calibration we should shoot for a 700 curve... I don't know, I have not calibrated mine... All measurements were with default White Balance and CMS.

If I was to make an informed guess... I would think that the C8 is optimal with 1000/1000/400 content and tone maps to a 700 curve... If indeed 700 is correct. But that 700 is not Peak Luminance.

QB

[SamuriHL]

That seems logical. But right now we can't really separate them. I don't calibrate HDR or DV specifically on my C8 because it's pointless to even try. I do have my white point set in the service menu to D65, though, and I use that for everything, including my calibrated SDR mode. In any case, I still think Panasonic chose 1000 nits for OLED for a reason and right now I am very happy with what I'm seeing by setting that as my DPL. Yes, being able to set 700 nits and force the metadata to 1000/1000/400 probably would be best but we can't do that right now.

Back to DTN for a moment, I think we absolutely DO need it. I just tried a 10000 nit video without it and it was an abject mess. As soon as I turned it back on and set it to 150 things are back to what I expect.

[QBhd]

I am guessing that is a 10,000 Nit demo video? Anywhere I can get a copy?

QB

[SamuriHL]

It's from the spears and munsill uhd demo disc. So you'd have to buy it but it's a very useful disc.

Sent from my SM-G975U using Tapatalk

[bkrieger]

See my responses in bold :)

I was able to get it to play smooth by setting the GPU to RGB Full, and the LG TV to PC Mode. Just want to verify, should MADVR be set to BT 7.09 or BT 2020 for both media AMD gaming? Also what should the sharpen edges screenbe set at?

Thanks

[VBB]

Is PC mode still locked to wide gamut? If so, you'll want to set madVR to DCI-P3, because that's the closest to native. Are you talking about the edge sharpening setting on the TV? I have it disabled where possible.

[chros]

Again... That's probably a 10% Window... Not the Peak that madVR will spit out... Super small super bright highlights will be capped to 700 well below what the display is capable of. That's why the measured 10% Window is actually brighter than 700 (730ish iirc)... I would guess that is the TV trying to pump up brightness to what it thinks is best.

Yes, I agree. So 700 is maybe 10% window on C8, but 1% window on B8. Let's say B8 can only do 200 nits less. So, if 1% is 700 on B8 then highlights maxed out at 800? That means 1000 for C8, and you don't have to care about "wrong" metadata.

Possible problems:
- we don't know the exact highlight peak value, that effectively means lieing to madvr :)
- what about movies like The Meg with really high max and avgFall? (I don't remember the exact values now) This can trigger ABL then...

[QBhd]

I wouldn't say using 1000 nits is lying to madVR... RTings value of 979 for a 2% Window Peak is pretty damn close to 1000

And just for reference the B8 managed 722 in the same test on RTings.

I am not saying that RTings is the gospel for such info, but they do have a long track record and take their methodology seriously... Let's call them a reliable reference

It's from the spears and munsill uhd demo disc. So you'd have to buy it but it's a very useful disc.

I actually think I might have a copy of that...

QB

[bkrieger]

Is PC mode still locked to wide gamut? If so, you'll want to set madVR to DCI-P3, because that's the closest to native. Are you talking about the edge sharpening setting on the TV? I have it disabled where possible.

How do I know if PC Mode is locked to wide gamut?
No, the upscale refinement screen. Should sharpen edges, crispen edges, and thin edges be checked and if so what should the numbers be?

Under properties in Madvr leave it on PC Level 0-255 and 10 bit?


Also when playing a movie when I hit Control J it says glitch 1. Any way to figure out what is causing that and get rid of it?

Still getting dropped frames but better than it was. I put the Madvr settings on low, Chroma Upscaling, Image Upscaling and image downscaling all on cubic 60

Thanks

[chros]

I wouldn't say using 1000 nits is lying to madVR... RTings value of 979 for a 2% Window Peak is pretty damn close to 1000

And just for reference the B8 managed 722 in the same test on RTings.

I am not saying that RTings is the gospel for such info, but they do have a long track record and take their methodology seriously... Let's call them a reliable reference
I agree but there are panel (settings?) diferences since my B8 only reaches 700 in 1% window using PC mode Full RGB. Anyway, normally everyone should measure his.

[QBhd]

What tools do you use to measure the 1% window (meter, software, panel settings, etc.)? I can't get a reproduceable result with my tools...

QB

[SamuriHL]

I may try to see what I can do with colourspace this weekend. Using my raspberry pi as the tpg I can force HDR using my Integral 2 so in theory I should be able to test 1% without even needing to touch the service menu. Maybe. :)

(Of course, once i set all that up I'm going to want to recalibrate the damn thing for SDR...it's been several hundred hours since I last calibrated it)

[QBhd]

I put 6.2 hours per day on mine... It's probably already over 3800 hours.

BTW, just to make sure we are on the same page, I use Cinema preset

QB

[SamuriHL]

Which will still be different for me because I've calibrated the D65 white point in the SM and use that for all my modes. So even though I'm using cinema I'm using a D65 white point.

[VBB]

How do I know if PC Mode is locked to wide gamut?

In the TV settings, Picture -> Picture Mode Settings -> Expert Controls -> Color Gamut. If it's grayed out and set to wide, it's locked.

No, the upscale refinement screen. Should sharpen edges, crispen edges, and thin edges be checked and if so what should the numbers be?

Entirely a personal preference. You'll have to judge for yourself. I don't use any refinements.

Under properties in Madvr leave it on PC Level 0-255 and 10 bit?

If your video card is set to RGB full and the TV is set to black level high, then yes. 10-bit is up to you. I would recommend going with 8-bit due to reduced banding.

Also when playing a movie when I hit Control J it says glitch 1. Any way to figure out what is causing that and get rid of it?

When you first start playback, there will be drops/glitches. That's perfectly normal. Give it a few seconds, then hit CTRL+R to reset.

Still getting dropped frames but better than it was. I put the Madvr settings on low, Chroma Upscaling, Image Upscaling and image downscaling all on cubic 60

You could also try setting the LAV decoder to "none", which is software decoding using the CPU. Maybe that will improve performance in your case.

Thanks

Hope that helps :)

[SamuriHL]

I did a little metadata testing. This is on a movie I'll be watching tonight. I wanted to compare the HDR optimizer on my UB820 to what madvr is sending.

madvr:
LUM 1000/0.005 maxCLL/FALL 1000/333

ub820:
LUM 1000/0.005 maxCLL/FALL 1000/0

passthrough:
LUM 1000/0.005 maxCLL/FALL 0/0

Admittedly not super interesting on a movie that:
min: 0.0050 cd/m2, max: 1000 cd/m2

So I pulled out the good ole Spears and Munsill demo disc. That's got a variety of fun content to play with. I decided to play with the 4000 nit BT.2020 HDR10 demo video.

madvr:
LUM 1000/0.005 maxCLL/FALL 1000/333

ub820:
LUM 1000/0.005 maxCLL/FALL 1000/800

passthrough:
LUM 4000/0.005 maxCLL/FALL 4000/1616

A little more interesting than my movie I suppose:
Mastering display luminance : min: 0.0050 cd/m2, max: 4000 cd/m2
Maximum Content Light Level : 4000 cd/m2
Maximum Frame-Average Light Level : 1616 cd/m2


I'm not sure exactly what this tells us but clearly the UB820 is also screwing around with the metadata that it's passing to the OLED display during it's tone mapping algo, as well. It seems to be chasing a higher FALL than madvr does though. I forget how madvr calculates the FALL. Anyway, I thought I'd post this in case it gives anyone food for thought. I simply found it an interesting exercise.

[QBhd]

I think it was chros who recently mentioned that madVR outputs MDL=DPL, Max CLL=DPL, Max FALL=DPL/3

I don't think LG cares about Max FALL... But your findings again point to 1000/1000 (MDL/Max CLL) as being optimal metadata for these devices.

QB

[SamuriHL]

I would agree with that.

Also I just set up for calibration (what a pain in my case but anyway...will probably do SDR while I'm at it). I took several readings of L10 which I believe is a 1% window size in Colourspace with metadata sent as 1000/1000/400 and watched it go up to ~890 or so. It does start to fight with me if I leave it like that for too long and starts dropping again. I'm not sure if I'm exactly measuring this correctly, but it seems reasonable to me. Nonetheless, I would definitely argue that you are correct about specular highlights getting into the 900+ range. Which again, points to metadata 1000/1000 being ideal.

[bkrieger]

http://https://share.icloud.com/photos/0WNACTrK5UNBY1MAgkxd4b__QWhat do you have your LAV decoder options set to? Choice of color space should have zero impact on performance. I've experienced the green screen with YCbCr, so I would avoid that. Could be an issue with only the 9-series video cards and newer drivers.

I had seen an updated beta Madvr I believe it was version 113 that I downloaded the other day. I did not release but that is what caused videos to be unwatchable. I am now back to version 92.17 and is back to how it was before.https://share.icloud.com/photos/0lJX_jZdFr9dgd261fbInrNIwhttps://imgur.com/gallery/t7VanKo

Here are my Lav settings



https://imgur.com/gallery/t7VanKo


https://share.icloud.com/photos/0-9XJENHC41przubaXt8-mShQ

[bkrieger]

Hope that helps :)

The LG Is locked to Wide Gamut using PC Mode

Thank You for all the settings tips and advice. I will try those and see if it helps.

I appreciate it.

Thanks

[chros]

What tools do you use to measure the 1% window (meter, software, panel settings, etc.)? I can't get a reproduceable result with my tools...
What do you mean it is not reliable? +/-10 nits in hdr is not an issue. :)
I have the same tools as you, nothing special, I've listed in that B8 post :)
Obviously the panel is pretty unstable in HDR. I used the settings that are documented in that post, and even with that bfi setting, the grayscale patches heat up the display, let alone if you leave the 100% white there just for 30.seconds, the peak nit is rising slowly :)
BTW, just to make sure we are on the same page, I use Cinema preset
I used that as well.
Which will still be different for me because I've calibrated the D65 white point in the SM and use that for all my modes. So even though I'm using cinema I'm using a D65 white point.
Me too, but can't we set this in HDR10 Cinema preset? If we can then I don't think it matters much.
Let alone using 100/120 nits SDR D65 for 203 nit HDR diffuse white... :) How does this even work? :) I guess if LS/CS gurus don't have an issue with it then it's good to go, right?! :)
I think it was chros who recently mentioned that madVR outputs MDL=DPL, Max CLL=DPL, Max FALL=DPL/3

I don't think LG cares about Max FALL...
I asked madshi about the calculation around a year ago, before I wrote that B8 post.
Yep, dunno about maxFall, but maybe LG uses it for ABL algos, who knows :)
I took several readings of L10 which I believe is a 1% window size in Colourspace with metadata sent as 1000/1000/400 and watched it go up to ~890 or so.
That's about right with the C8 (+~200 than B8). I told you guys to measure your C8s back in the day! :)
What sad is that newer CX models has almost the same peak as my B8 :) The C8 is the brightest out of all so far (except for the special Panasonic editions, gz2000/etc).

[QBhd]

Hmm... I actually had to look that up... it seemed odd that the C9 and CX would be less bright than the C8... but yeah a quick peek at RTings does confirm that... Wow

Not that I didn't trust you... but I have always been a guy that fact checks ;)

QB

[Asmodian]

it seemed odd that the C9 and CX would be less bright than the C8... but yeah a quick peek at RTings does confirm that... Wow

I noticed the trend downward even when calibrating 100 nits SDR, moving from the C7 to C9 to CX, each one needed a slightly higher OLED light to hit 100 nits after calibration.

I assume it is due to concern over burn in.

[QBhd]

I think you have to have some seriously bad viewing habits for burn-in to happen... I am approaching 3800 hours on my C8 (6.2 hrs/day) and I don't have any burn-in. Plus I think I like my SDR brighter than most (OLED Light is at 56)

QB

[chros]

Not that I didn't trust you... but I have always been a guy that fact checks ;)
:) No worries, I can also remember wrong.
I think you have to have some seriously bad viewing habits for burn-in to happen... I am approaching 3800 hours on my C8 (6.2 hrs/day) and I don't have any burn-in. Plus I think I like my SDR brighter than most (OLED Light is at 56)
Yes but LG is concerned about the average Joe not us :)

[SamuriHL]

Me too, but can't we set this in HDR10 Cinema preset? If we can then I don't think it matters much.
Let alone using 100/120 nits SDR D65 for 203 nit HDR diffuse white... :) How does this even work? :) I guess if LS/CS gurus don't have an issue with it then it's good to go, right?! :)


There's been a lot of debate about white point in HDR as you know. I'm not knowledgeable enough to argue the merits of either or. In any case, yes it can be set very easily because you just set the color temp to the one you calibrated. In fact, ConnecTEDDD recommends calibrating 2 different color temps in the SM, one for SDR and one for HDR. I've not done that yet but that's how you can get your 203 nit HDR white if you so choose.


That's about right with the C8 (+~200 than B8). I told you guys to measure your C8s back in the day! :)
What sad is that newer CX models has almost the same peak as my B8 :) The C8 is the brightest out of all so far (except for the special Panasonic editions, gz2000/etc).

I did measure it back in the day and it was lower. I was also using Calmann back then and attempting to calibrate HDR and DV. Which is futile on these panels. Simply calibrating white point is good enough for HDR and DV. Obviously I have a 3dlut for SDR and I do not touch the 1dlut at all (stupid LG bugs!). But yea, the C8 is quite the bright display which I love. I can see very clearly on some of my test content that by setting DPL to 700 we were indeed limiting the overall specular highlight brightness.

I think it's time for a LOTR viewing...

[SamuriHL]

I think you have to have some seriously bad viewing habits for burn-in to happen... I am approaching 3800 hours on my C8 (6.2 hrs/day) and I don't have any burn-in. Plus I think I like my SDR brighter than most (OLED Light is at 56)

QB

I come from plasma so burn-in avoidance is in my blood. And, uh, yea, that's quite bright for SDR. Mine's set at 30. :)

[QBhd]

My living room is not super bright, but it is very far from dim... RTings recommended 56 for bright room and when I finished testing various things it worked the best for my taste. I have a triple lamp in a far corner pointed at the wall and ceiling... I don't watch SDR movies anymore and for TV shows, OTA, and YouTube it's good enough... When watching an HDR movie I will turn one or two of the lights off... Dolby Vision is black out, and wow does it look amazing.

I too came from Plasma, so yes B-I avoidance is always in the back if my mind

QB

[VBB]

I don't watch SDR movies anymore

:eek::eek::eek: :D

Nothing wrong with a brighter SDR picture. I settled on 46 years ago.

[SamuriHL]

Blasphemers! :D LMAO! No I get it. I have a very bright room in the day time but I do most of my watching at night.

I gotta say, Fellowship of the Ring is damn impressive. Yes, I could have watched it in DV but I wanted to see what HDR would do with my new settings.

[chros]

Yes, I could have watched it in DV but I wanted to see what HDR would do with my new settings.
Herecy! You don't respect the director's intent either! :D

[SamuriHL]

Well if they'd fix madvr I would! :p :D

[SamuriHL]

I did some recalibration this weekend. I created a white point for 203 nit to use for HDR. I need to do some more testing on that. At first glance the results are interesting. I'm not sure they're correct, though.

I also changed my SDR calibration to DCI P3 D65. I've come to the realization that this display is 95% DCI P3 and that all of the source I play on this display are capable of sending that. Some quick tests with madvr set to DCI P3 calibrated display and the bt.2020 checkbox set looks really good. I'll be playing around more this week but so far this has been fun.

[QBhd]

Is that 203 Nit white point really a good idea? When I Google HDR white point it still says 100ish (the 93.blahblah of SDR)... I am still in the process of learning calibration, that's why I just measured for now.

QB

[SamuriHL]

Probably not, no. There's a whole huge debate on this topic. Most UHD's are mastered around 100 nit white point while supposedly newer ones are using a 203 nit white point. Maybe. I did it just to see what the result would be. It's super easy for me to switch between them (and because I don't create a 3d LUT for HDR it hardly matters) for comparison. Which is why I did it.

Do I think it's a "good idea?" No, not really.

[QBhd]

Okay... so I pulled out my S&M UHD disc... and that 10,000 Nit demo is interesting to play around with DTN. I am not sure DTN is really the best way to describe this feature of madVR. This demo has tons of excellent scenes with varying FALL and CLL. So when you get to the horses in the snow it is very evident that DTN is needed. So I did a test of DTN=1 and DTN=500 (the highest you can set it) I have two profiles for HDR and set-up a short cut to switch between the two. 500 made the horse scene come to life, 1 basically did nothing.... anyway, I used this technique to find out that all you need to make the horse scene awesome was a DTN of 10... okay that's interesting... but the scene a few before that with the sun shining through a valley on the top of the mountains where the peaks are bright and the rocks are shadowed a little... this was still overblown and the contrast was not there, so I did the switching back and forth for a bit and found that 50 DTN brought that scene to life (without changing the horses).

This value seemed good since it was not dimming the low nit scenes, yet giving those scenes that needed it room to show the range... but then I got to that scene with the buffalo on a yellowish field... the yellow is still overblown here... so 75 DTN... not perfect, but it maintains excellent range on the other two scenes.

I think Dynamic Target Nits is just a wrong description... it's more like it is giving more room to the dynamic range, and it's not a nit value per say but more like a percentage... especially if you pull up the detailed HDR info on the OSD.

Anyway, I did all this with no compr limit set at 0, ... blah... let's just do a SS:
https://bn1301files.storage.live.com/y4mof0PbSxmakrmgKbBu81vG7n6R9CIvTdf5ONlaPScot2UlnVBROtuOV9Myljy5xzOdVx7p939OJRyx1wrYF5Hv0TAv3Db3ABVpSvHR3aWW9iuJ6ZCNTTbsLEa8vMfDC2mRVEsHHEfeYQfmA8BgbkIjP45TaZl2lDC-jjKFjADetvOmpCUMFKa_QebpF1NJ0kD?width=826&height=410&cropmode=none

So if I get a chance later tonight, I will explore what no compr limit actually does.

QB

[SamuriHL]

Yes, it IS a percentage. It allows some scenes to go above your DPL. Which is very useful for us. That's why I said we need it. I truly believe that if we can set the DPL accurately, metadata to 1000/1000/400, and dial in the DTN we'll be in heaven with these displays. I haven't had time to redial it in. But those scenes, yes, that's why I use them. To evaluate madvr tone mapping. Because at 10000 nits, you're getting some serious compression. And because he includes lower nit versions, you can compare to what it should look like. This is where the fun begins. :)

[QBhd]

I've been thinking about the behavior I saw paired with the measurements I have already taken and I don't think it's about going above the DPL. It's fitting more range into the DPL window. If you have DTN off (or set to 1, which is basically off) anything above 1000 (the current theory's best MDL for the C8, and probably really close to it's actual Peak Luminance) is basically clipped and squeezed at 1000. With a 10k source that is almost all white is displayed the same and it looks like hot garbage, literally... but the scenes that are below 1k all look great. If we slowly increase DTN, more range is allowed into out 1000 nit window and as such we see more detail in those blown out whites... however there is a cost, those darker scenes start to dim since they are now being squeezed too. With my quick and dirty testing last night I came to the conclusion we do need some, but we also don't need much. For example the horses in the snow only needed 10, anything above 10 had no perceptible effect on that scene (I think it was 10, if not, it was shockingly low to what I expected). There is a balance that must be found. Plus there is the "no compr limit" that needs to be explored (next on my to do list) to see what that is actually doing with a 1000 nit "window"... I am pretty sure what that will be doing, but the engineer in me needs to test and verify to be sure... and who knows, maybe it too does things that are not entirely evident in the name.

DTN is a terrible name... I always thought it was a target nit that madVR would shoot for. Like we have 1000 DPL set and if we set DTN to 200 we would have an image that is around 200 nits. Since I like a brighter SDR, it made sense to "target" a brighter overall HDR... but that is not really what I see with a 200 DTN. It's more like there is 200% more range to be used, and this really does a number on the lower nit scenes... anyway, terrible name :P

QB

[SamuriHL]

Agree on the name. It has confused nearly everyone.

My understanding (which very well could be wrong) is that DTN will allow a frame to go above DPL in some cases. My thoughts on exploiting this is to set things up like this:

DPL: Measure 10% window nit level (e.g. 750 or whatever)
metadata: 1000/1000/400
DTN: tbd but enough to get us to 1000 nit on 10000 nit scenes

If we can have this so that highlights can go up to the theoretical 1000 nits that our chosen curve allows (and we know the 1% window on the C8 can nearly accommodate) then madvr should tone map everything to the DPL while allowing highlights to go above it, which is exactly what we're trying to achieve. But this is all complete theory right now. It's just my wish.

[quietvoid]

I don't think DTN is supposed to be affecting scenes with a peak under the DPL, at least from previous testing.
Increasing it does increase contrast and highlight detail, at the cost of making the rest of the picture darker.
In the horses in the snow scene, this is obvious when comparing to passthrough. The horses lose all their color and just look dark.

Maybe setting a high no compresssion limit would help avoid making things too dark in those very high brightness scenes, never tested.

[SamuriHL]

I've not tried that, either. It would be a good test to try.

[chros]

QBhd, you are right about what DTN does. I use 55 (I don't remember why :) ), it makes the image a bit brighter when it's in effect.

But my biggest question is this: what does or how does behave the LG internal 700 nits target?! I can't test it, since my peak is around that.
Maybe you guys can take a look, with content between 700 and 1000 nits and setting madvr dpl to 700 and 1000. (but 700 results in brighter image on LG). Because if you set 1000 but LG is clipping at 700...

[SamuriHL]

That's why I very badly want to be able to set DPL to a measured real display nits and then use the metadata to force it into the proper tone curve. Technically speaking I can PROBABLY do this already. I may try here in a minute.

For me I'm really liking what I'm seeing at the moment with:

https://i.ibb.co/Z6y8L9f/madvr-2020-12-08-17-23-54.png (https://ibb.co/qMbsy6D)

[chros]

And guys please take a look at this post (https://www.avsforum.com/threads/improving-madvr-hdr-to-sdr-mapping-for-projector.2954506/post-60278979)!
I wrote about this 2 times already and literally no one gave the slightest FAQ :) That's a big bug especially for us.
If you agree then quote that message and write madshi (preferably before the next beta) otherwise it will never be fixed :)

It should be only 2 "if"s:
- don't do any clipping if measured frame peak is lower than DPL
- if it's higher then only reduce frame peak nit to DPL (and not below it!)

[quietvoid]

Dynamic clipping is essentially a way to counter low nits titles in SDR, they usually end up flat otherwise.
Examples IIRC are Game of Thrones S01, Murder on the Orient Express (2017).
I think the same overall brightness uptick can be applied to HDR, however it goes against the coded PQ intended value.

It also helps in weird content like Mad Max and that Harry Potter film, where you would have grey-ish highlights if they were not clipped.

[chros]

I think it's useful for us as well (Mad Max Fury road, etc).
But even if it isn't, a bug is a bug! :)

[quietvoid]

I'm not sure if it is a bug, though I have no idea what the algorithm is and when it applies.
It could apply to both scenes lower and higher than DPL for all I know.