MadVR in use with LG OLED Thread

[SamuriHL]

What is the DTN for 900?

[aron7awol]

Originally Posted by SamuriHL

What is the DTN for 900?

Since we can only do integer DTNs, I've been choosing DPLs that end up with the exact same target. So in that first set of values, I chose 897 instead of 900.

I'm happy to throw some combos out for you if you give me your "default" DPL and DTN combo you want to compare to.

Don't worry about the fact that the DPLs don't necessarily end up exactly where you want them. After you do the apples-to-apples comparison and see the relative differences and hone in on which you like best, we can then do further tweaking to get to the exact DPL you really want.

As an example, the perceived difference between my personal 875/80 and 805/81 combos is so miniscule that there's nothing detectable in ~99% of frames and only those particular special frames I was able to find can I even detect the tiniest bit of clipping difference! So really, don't worry about the exact DPLs for this phase of testing!

[SamuriHL]

I don't have a great combo right now as I've been tweaking the crap out of it since these new builds have come out. Bouncing between a higher value (150) to lower values all the way down to 65. None of them really seem great in all content so far for me. But this was at 1000 dpl. Of course I've changed the dpl to 900 so that will almost definitely require going through content again. I was starting to lean towards somewhere around 75 ish the last time I messed around with it. God this is maddening. LOL

[aron7awol]

Originally Posted by SamuriHL

I don't have a great combo right now as I've been tweaking the crap out of it since these new builds have come out. Bouncing between a higher value (150) to lower values all the way down to 65. None of them really seem great in all content so far for me. But this was at 1000 dpl. Of course I've changed the dpl to 900 so that will almost definitely require going through content again. I was starting to lean towards somewhere around 75 ish the last time I messed around with it. God this is maddening. LOL

It doesn't really matter. Just pick one that seemed pretty good (75, 100, whatever) for now and then compare some equivalent combos to that. The exact DTN won't matter, the comparison will show the DPL differences only. Use this to hone in on DPL then worry about DTN later

[SamuriHL]

I see what you're saying. I was doing it backwards....yea that's a good idea. I'll see if I can find a few minutes tonight to at least give a cursory run through. The C8 should be good close to 900 DPL.

[aron7awol]

Here are the three formulas I use (always assuming "don't add peak nits" is checked):

DTN = multiplier * 50 / log10(DPL)
multiplier = log10(DPL) * DTN / 50
DPL = 10^(multiplier * 50 / DTN)


So if we pick 1000/75 as your "default", we get 913/76 and 836/77 as equivalent combos.

My guess is if you compare those 3 combos on the right specific frames, you'll perceive the tiniest clipping difference between 1000/75 and 913/76, and maybe the tiniest brightness difference between 913/76 and 836/77.

Edit: For anyone who wants to calculate these equivalent combos themselves, use the second equation to find the multiplier for your "default" DPL and DTN. Then increase/decrease DTN by 1 and then use the third equation to find the DPL for that DTN and the same multiplier you just found.

[quietvoid]

Okay, I had a quick look. Compared 1000/125 and 897/127 and got to the same conclusion. 1000 is clipping even though madVR has the same tonemap target for both.



Next is to see how high I can increase DPL before it becomes worse.

[aron7awol]

Originally Posted by quietvoid

Okay, I had a quick look. Compared 1000/125 and 897/127 and got to the same conclusion. 1000 is clipping even though madVR has the same tonemap target for both.

Next is to see how high I can increase DPL before it becomes worse.

Well this is good, and expected, but at least now the only difference you should be seeing is that clipping and not an overall brightness change of the image (other than where it is clipping, which may be quite a bit in that particular one) due to a different curve being used.

I think the next logical steps would be to compare the 897/127 to 851/128 to see if there is another perceivable reduction in clipping, and also compare it to 947/126 to see if there is a perceivable increase in clipping.

[quietvoid]

Thanks. I did what you suggested and 851/128 brings a little more detail, while barely dimming.
947/126 is clipping just like 1000/125, and there's no noticeable improvement from 807/129.

Since I've switched to 100 DTN, I'll probably settle for 841/101.
No clipping and there's not much dimming other than those extreme FALL scenes (really, the horses one is the worst).
It's minimal and would rarely happen in real content.

For those who want to keep as much contrast, 851/128 would be it.
I think it makes The Meg skies a bit too dim and they look unnaturally saturated then, but anyone's free to want this.

I've looked at the clip you shared from BvS about saturation, I still prefer using desat 4 on top of desat 1, otherwise the lightning around the giant is really orange and looks dim.
Though I've lowered it to 100/100. Didn't try those low brightness shots with faces, but I'd assume those are no problem for us.

These are gonna be my (hopefully) final settings until madshi changes everything again: https://ibb.co/2YMt0tD
At least we've nailed DPL/DTN

[aron7awol]

Okay, so you found an improvement from 897 to 851, but not from 851 to 807. So your "optimal" DPL is probably somewhere between 851 and 897. Now if you want to go back to ~100 DTN, don't worry about using one of those exact combos, like 841/101. You can just use 850/100 or 875/100 or whatever you want, the difference is so small that now that we aren't directly comparing for clipping it doesn't matter. No reason to leave those 10 nits on the table Then you can go back to experimenting with just DTN now that you figured out your approximate optimal clipping point.

[quietvoid]

That's true, I'll have to see if it's worthwhile.

[quietvoid]

Well I compared 850 up to 880, there's no noticeable brightness increase.
Clipping starts to be noticeable at 880 DPL, but even then it's not something you'd notice when actually watching.
So it doesn't hurt me to set it at 870.

[SamuriHL]

That seems a reasonable DPL for these panels.

[SamuriHL]

Yea I'm liking that a LOT. Like a WHOLE lot. That fixes my crocodile issues in SM. The green spear looks fantastic. I've got DTN set to 127 at the moment and will play around with that more but 870 for the C8 is a very good DPL. Max lum, highlight sat I adopted your infinite (really unsure on this one), apply 2-4 100/100 is what I've been using and see no reason to change it. I did up my repair range from 0.010 to 0.020 but have no idea if that really matters. ceiling multiplier I still like at 4x. Yea, this is very good!

[quietvoid]

Repair range doesn't seem to do much (or anything) with max, it's noticeable with sep though.

[SamuriHL]

There's no reason for us to use sep that I've seen. I think we're pretty well dialed in. Obviously a little tweaking here and there but big questions have been answered.

Sent from my SM-G975U using Tapatalk

[aron7awol]

I think sep (or anything other than max) causes artifacts when used in combination with the non-defeatable TMing on the displays. At least it does for me on the green spear. I think maybe the two curves interacting end up causing a kink in the resulting curve or something. So it's max only for me, unless someone can point to an actual artifact caused by it. Every "artifact" I've seen has just been an accurate representation of the source.

[SamuriHL]

Very likely. As long as we can keep the options we have now we're in good shape.

Sent from my SM-G975U using Tapatalk

[aron7awol]

Yeah, I'm pretty happy with my current settings, too. The tunable ceiling makes a huge difference on The Meg. I still can't help but feel like the DTN/ceiling logic can still be optimized quite a bit more, but I actually think it would take someone taking the time to compile a whole spreadsheet of many different frames - FALL, FALL stddev, avgHL, and desired target (through experimentation on each frame) and then we'd have a shot at creating an algo to hit that whole range as closely as possible, but we might discover at that point we can do even better with some additional frame data. It's pretty good as-is, but I still think the target isn't quite where we want it in a lot of cases. Just comparing The Meg and S&M 10,000 there's a huge difference in the DTN and ceiling combo I like on each.

I'd actually be really interested in seeing what a neural network would come up with if fed all the different grades of the S&M demo with the goal of trying to create each of the lower grades from each of the higher grades using madVR Unfortunately it's one of the only things we have in so many different grades of the same exact content, but it does at least have a decent mix of scenes.

[SamuriHL]

That would be an awesome idea. I suspect they're going to focus on the PJ stuff a lot more from here but at least we got our wish for a few things that helped us a lot. I would argue that he should change the multiplier (since he only wants 1) to 4x or maybe even 5x. We'll have to tell him soon what we want there cause he doesn't want to leave the selectable option in place.