madVR - high quality video renderer (GPU assisted)

[huhn]

when a simple scaler is used madVR does seem to do 420->422 then 422->444 so you can ignore the line. there is no issue here.

if you disable output formats and lav send RGB then you can not upscale chroma because there is no chroma.

nothing of this has anything to do with what your device does or what madVR outputs. madVR output is RGB always,

[djsolidsnake86]

i don't understand if is better use d3d11 or dxva copy in lav filter?

[huhn]

both have the same quality.
d3d11 is faster but quite a number of features don't work.

a copyback decoder is slower but all features work.

[el Filou]

Originally Posted by SirMaster

This is what I use on my projector that is about 85 nit peak white. https://nicko88.com/misc/madVR%20set...%20-%20hdr.png I quite like the effect of putting a lower peak nits and using the no compression limit to compensate. This helps increase the effect and the effective range of contrast recovery while still using a lower strength curve option.

But you're wasting 35 nits of your projector's dynamic range by doing that.
madshi has stated numerous times in the AVS thread it's a bad idea to not set the peak nits setting to the actual real peak nits you have. It's better to set the real peak nits and use other settings to get an image you like.

[Sunspark]

Originally Posted by huhn

both have the same quality.
d3d11 is faster but quite a number of features don't work.

a copyback decoder is slower but all features work.

Power consumption will vary depending on your system. For example, on mine using madvr overlay mode with dxva2-native will cause fan whine due to the increased load caused by the need to convert the texture format to another format. Paradoxically, copy-back will have less load in this circumstance because it is not doing a conversion even though there is more bus traffic.

Only way to tell other than hearing the fan, is to use something like gpu-z to see the load in various configurations, and also how much CPU is being used for any given combination.

[SirMaster]

Originally Posted by el Filou

But you're wasting 35 nits of your projector's dynamic range by doing that.
madshi has stated numerous times in the AVS thread it's a bad idea to not set the peak nits setting to the actual real peak nits you have. It's better to set the real peak nits and use other settings to get an image you like.

Other settings like what exactly?

I have been using madVR for over a decade and madVR tone-mapping for as long as it's existed in madVR.

I have experimented with all the settings for hundreds of hours and these are the settings that I like the look of the most.

Setting the DLP lower makes the highlights brighter. And setting NCL > DLP stretches out the dynamic range.

Also I love HSTM, but it's not active above DPL. Setting DPL lower allows HSTM to be active lower and I like what that's doing to the darker scenes. If only we could just activate HSTM for the entire range.

To me, with these settings the image subjectively has more dynamic range than with a higher DLP rather than less.

I have another preset with 85 nits set for DPL, and I toggle it all the time, but I never prefer it over my other settings.

I can take a screenshot comparison tonight to show what I mean about why I like my settings.

[70MM]

Many thanks for your explanation SirMaster, I for one would love to see your comparison screenshot.
What I find difficult for begginers is there are so many settings in HDR to SDR that one needs to be really competent to know, or see what works best.
All some of us can do is to listen to experienced users like SirMaster and try their settings.

[huhn]

the options are mostly for testing not for actual use. they where never meant to stay like this.

if you want your system to work properly i would suggest to avoid beta versions.
there are known flicker issue(mostly fixed as far as i know in the newest betas) and other not yet removed issues.

[70MM]

Funny thing Ive never ever seen any flicker issue.....So do you think the Tone mapping on the JVC projectors is any better than using 113?

[Alexkral]

@el Filou

I do this most of the time with my lcd display. The result is expanded to the actual peak nits, so there is no wasted dynamic range.

@SirMaster

I'm curious how this actually works. The "no compression limit" parameter seems to mean that the content remains untouched below it. To achieve this, it's not possible to use a display peak luminance equal to 50. So it seems to me that in this situation, a display peak luminance above that is chosen (say at 150 nits), which guarantees that content below the compression limit is not actually compressed. But then the result must be scaled down to 50 nits, and this final result is expanded to 85 nits by your projector.

For brighter highlights, the original BT.2390 roll-off would be much simpler if it were available.

Edit: I don't know how the result can be scaled down, unless what is sent is not in the full range [0,1]. Can you check if setting a display peak nits higher makes the result different with the rest of your settings? I would say that in this situation it is not used at all.

[flossy_cake]

Regarding power consumption, I'm finding some decent power savings on my GTX 1070 by setting "adaptive" power profile for the media player's process -- in my case mpc-hc64.exe. The amount of saving appears to fluctuate depending on the scene though -- on some scenes I'm getting as much as 40 watts savings, but on a typical scene like 1080p Star Trek episode it's about 20-25 watts savings.

Also tried enabling CPU downclocking & core parking through Windows power settings and found that could save an extra 5 watts but I didn't think it was worth it (i7-4790k).

Finally saved about 100 watts by setting the screen saver to a blank screen and set the taskbar to "auto hide" to prevent burn in.

[SirMaster]

Originally Posted by Alexkral

@SirMaster

I'm curious how this actually works. The "no compression limit" parameter seems to mean that the content remains untouched below it. To achieve this, it's not possible to use a display peak luminance equal to 50. So it seems to me that in this situation, a display peak luminance above that is chosen (say at 150 nits), which guarantees that content below the compression limit is not actually compressed. But then the result must be scaled down to 50 nits, and this final result is expanded to 85 nits by your projector.

For brighter highlights, the original BT.2390 roll-off would be much simpler if it were available.

Edit: I don't know how the result can be scaled down, unless what is sent is not in the full range [0,1]. Can you check if setting a display peak nits higher makes the result different with the rest of your settings? I would say that in this situation it is not used at all.

Setting a higher peak definitely affects the picture. If I set a higher peak, pretty much everything gets dimmer.

NCL is inherently set to DPL, or rather, can't be lower than DPL. So if I put NCL to 0, it would really be 50. But setting it to 85 causes bright things to get brighter, and yet keeps the dark things darker which to my eye increases the dynamic look of the image.

It also allows HSTM to work down to 50 nits rather than down to the real DPL which would be higher.

This makes HSTM have a larger effect. But I still use the conservative HSTM curves and never see any flicker because I am using the the conservative curves that don't seem to cause flicker.

I don't know how the algorithms work or how they combine in this case, but I would say this is not at all an intended way to use these controls. But the point is after much experimentation I still like and even prefer them for my use.


On my 100 nit actual calibrated PC monitor, I often compare these 3:

DPL=100, NCL=0
DPL=50,NCL=0
DPL=50,NCL=100

I find I always prefer the last.

DPL=100 is the "correct setting".
DPL=50, NCL=0 just makes everything brighter and doesn't really look better. Near black gets too gray, etc as everything is rendered brighter than it's "supposed" to be.
DP=50, NCL=100 something "magic" happens. Near black, dark stuff stays dark like it should be, mid tones and bights get brighter like they are with DPL=50. Overall dynamic look is increased to my eye, with no noticeable downside.


I picked DPL 50 because I find that's the lowest I can take it before things start to "fall apart" or look flat, etc in various ways.

And I do keep DTN enabled and that also set at 50, so the target is not locked to 50, but uses the DTN algorithm.

So my conclusion has been, set DPL to 50, and NCL to real display nits. But this only scales to about 100 nits. If I have more than 100 nits, I will start to raise DPL then.

I just don't like how dark the image looks overall when the display has less than 100 nits, and you put in the real DPL which was the whole reason I played with the settings like this.


I don't claim this is correct or best for everyone, just that I know I like it. And friends who have similar setups I have showed them the comparison too and they have come to prefer it as well.


I didn't have a chance to get screenshots, but I will post some tonight.

[Sunspark]

Originally Posted by flossy_cake

Finally saved about 100 watts by setting the screen saver to a blank screen

This is a really stupid Windows 10 bug. I could hear the CPU fans whining while it was in screensaver mode. I checked the GPU load and it was spiked up high when the screensaver is active.. even ones that have been around for 25 years. Blank screen is the only one that doesn't spike the power now.

[Alexkral]

Originally Posted by SirMaster

I don't know how the algorithms work or how they combine in this case...

Me neither, but even though it's clear that HSTM is quite different from BT.2390, it doesn't seem like NCL is really doing what it's supposed to.

[SirMaster]

Originally Posted by Alexkral

Me neither, but even though it's clear that HSTM is quite different from BT.2390, it doesn't seem like NCL is really doing what it's supposed to.

Oh good point.

I know HSTM doesn't work below DPL, but it definitely works below NCL.

madshi did say he could make HSTM work down to 0. I can't remember or not if he actually enabled that option in one of the more recent builds.

Yep, found the changelog for it.

https://www.avsforum.com/threads/imp...#post-61351881

"3) Added contrast recovery option "with dark scenes, as well". By default, contrast recovery (HSTM) activates only if the frame peak is higher than the display peak. But this new option allows you to always turn HSTM on. Doing this will in some way go against director's intent (or at least against what the disc asks for), so I don't recommend using this new option. But you're the boss of your home theater, of course."

[70MM]

SirMaster do you set your video card to RGB Full 8bit or 10bit?
Also I see you tick under dithering Error Diffusion option 2 with "use colour noise" and "change dither for every frame" but both these add noise it says. Why do you "tick" these ( I have never ticked them due to add noise) and what benefits do they actually bring to the table. Many thanks...

[flossy_cake]

Dithering is an very cool and awesome technique that can make 8bit as good as 10-12bits, and the dither pattern ("noise") will be totally invisible if it is coloured subpixel dithering and temporally changing every frame. Even non-temporal, if done right, will be practically invisible. You can check this by taking a screenshot with and without dithering. If you open the screenshot in Photoshop and zoom right in and inspect pixel values you should see small fluctuations of +/- 1 steps depending on the pattern.

[Asmodian]

I prefer non-color dithering and not changing every frame. It is completely imperceptible to me. Changing every frame makes it more noticeable.

[SirMaster]

Originally Posted by Alexkral

Me neither, but even though it's clear that HSTM is quite different from BT.2390, it doesn't seem like NCL is really doing what it's supposed to.

Here is a simple screenshot comparison to show what I like about the DLP and NCL I use.

I am going to use a 100 DPL baseline assuming your display was 100 nit for example.

https://screenshotcomparison.com/comparison/25235

Here I am comparing:

Default: DPL100 - NCL0 (this is how most would recommend madVR, with correct DPL, NCL=DPL=100 by default here)
Mouseover: DPL50 - NCL0 (DPL half the true value, NCL=DPL=50 by default here)

The obvious result is everything gets brighter, and more compressed?, because DPL is lower of course. This doesn't look any higher dynamic range or anything. This is bad, and I wouldn't use DPL50 - NCL0 like this on a display with 100 real nits.



Now compare these:

https://screenshotcomparison.com/comparison/25236

Here I am comparing:

Default: DPL100 - NCL0 (same as before, the "correct DPL")
Mouseover: DPL50 - NCL100 (DPL50 like before, but this time increase NCL to 100 to match real display nits)

The result is the overall picture is a similar brightness, except now:

The darks are a bit darker (but not clipping or losing shadow detail to my eye).
The mids are about the same as before.
The brights are brighter (but not clipping either).

To me this has a look of a higher dynamic range, and almost like a "veil" is lifted.

So this is how I run my madVR. DPL=50, NCL=real display nits. And this works well in my opinion for displays between 50-100 nits (most projection setups).

For 120 nit projector I might consider something like DPL=(50 + (real nits - 100)) = 70, NCL=120

[SirMaster]

Originally Posted by 70MM

SirMaster do you set your video card to RGB Full 8bit or 10bit?
Also I see you tick under dithering Error Diffusion option 2 with "use colour noise" and "change dither for every frame" but both these add noise it says. Why do you "tick" these ( I have never ticked them due to add noise) and what benefits do they actually bring to the table. Many thanks...

I don't use change every frame anymore normally I looked at my HTPC and it's off.

But I used color dithering because it reduces luma noise (which is more noticeable than chroma noise) in theory?

I don't really think I could tell a different either way anyways.