madVR - high quality video renderer (GPU assisted)

[tp4tissue]

Originally Posted by 70MM

Ok thank you I understand, however that isnt an issue with me anyway as I have the 1080Ti card and using the highest NGU settings upscaling everything to 4K. My rendering hardly ever goes above 20 and I have no dropped or repeated frames for days/ever.

Luma scaler difference is the only thing that's visibly different (At a Distance).

For chroma, dithering, etc, the differences are really only visible if you walk in real close, or zoom in.


The sharpening algorithms are also very visible, but they're taste-based, and not a better or worse situation.


LOOKING FOR a difference is mostly academic, because you simply wouldn't notice much of it while a movie plays.

If knowing it's there helps you sleep at night or justify equipment purchases, then Zoom in, squint real hard, the little hairs on people's faces will look just a tad sharper.

[70MM]

thank you for the comments...

[Asmodian]

Originally Posted by 70MM

Can you explain to me please huhn why ordered dithering is better than error diffusion 1 & 2 please? Thank you kindly for helping...

Error diffusion is a tiny bit sharper and more random while ordered dithering has the lowest apparent noise. All three options are very good, I think I prefer ordered dithering on my 55" OLED TV while I like ED 1 or 2 on my LCD monitors, but the differences are also very subtle. Ordered is a great option because it is so high quality while also being fast.

[Warner306]

Originally Posted by Alexkral

The problem with HDR being too dark is still there though. 100 nits for diffuse white is too low in not ideal conditions (I'd say it's even too low in ideal conditions but that's just my oppinion) compared to where most people is used to see it in SDR. ITU-R BT.2408-0 recommends diffuse white = 203 nits for PQ and HLG production on a 1000 nits peak luminance display. Another study in the same report suggests diffuse white at 140 nits for indoor scenes and 425 nits for outdoor scenes, for PQ with 4000 nits peak luminance. I think I have read somewhere that at least some titles now are being graded with diffuse white at 203 nits. This should be the best way to solve this. We loose 1 stop at highlights but the overall brightness match better with SDR.

I wouldn't be against raising the value of reference white slightly for PQ HDR content. The problem is all existing HDR displays are tuned to tone map for a reference white of 100 nits, so I don't know how you would get around that without breaking support for all legacy HDR displays. I don't think HDR is too dark when viewed without ambient light in the room.

Displays could at least have a standard brightness slider like video games to make HDR a little brighter during the daytime, even if raised the bottom of the image a little.

Originally Posted by Alexkral

If I'm not wrong, both SMPTE ST 2094-10 (Dolby Vision) and SMPTE ST 2094-40 (HDR10+) curves change dynamically the knee point in a similar way that what Soulnight's utility with dynamic target nits is doing now with the BT.2390 EETF.

The knee point just changes where the tone mapping roll-off begins. The curve always slopes to the right away from the PQ curve, so the image never gets brighter; it only gets darker.

Soulnight's tool increases or decreases the target nits. Increasing the target nits raises the knee point further up the PQ curve to reduce compression, but it also changes the absolute brightness of all values relative to the output display. For example, if the display is 100 nits, selecting 200 nits will lead to a loss of more or less half of the calculated brightness of the tone mapped values because the target display is assumed to be twice as bright as the actual output display. The gamma curve is relative, so the output brightness isn't always that precise, but it is close.

When the target nits is higher, the pixel values are more spread out, but they use more of the lower region of the gamma curve to handle the majority of the pixels 0-100 nits. The pixels that represent the highlights are separated to the brightest end of the gamma curve. So the image has more contrast between bright and dark detail and looks less flat because the pixels are less compressed, but the overall image ends up darker.

[tp4tissue]

Originally Posted by Warner306

I wouldn't be against raising the value of reference white slightly for PQ HDR content. The problem is all existing HDR displays are tuned to tone map for a reference white of 100 nits, so I don't know how you would get around that without breaking support for all legacy HDR displays. I don't think HDR is too dark when viewed without ambient light in the room.

I noticed that calibrating to 2.2/2.4 (relative) gamma on LCD is much more similar in appearance to 2.2/2.4 (absolute) gamma (1886) on CRT/Plasma/OLED, vs using absolute gamma (1886) on LCD which makes everything greyish.

So, the near crush dark tones when using relative gamma on LCD is probably Correct. It's just more crush like in home environments where the room light is on.

The moral of the story is, it's suppose to look like that, just turn the lights off.

[SamuriHL]

Shockingly 418.82 didn't break HDR switching. LOL

[tp4tissue]

Weird Gamut behavior.

So on my ATI card, my spare bedroom tv reports a slightly wider Gamut +10% vs my Nvidia Card.

Is it because the Nvidia card is using the alpha channels differently ?

Does madvr use alpha channels in output or only 24bit.

[nevcairiel]

There is no alpha channel on the output. Different results might be up to dithering.

[tp4tissue]

Originally Posted by nevcairiel

There is no alpha channel on the output. Different results might be up to dithering.

Hrrrm..

I guess that must be the tv's gamut clamping software intervening with the nvidia output.

Is there a fundamental output difference between ATI and Nvidia, some sort of flags ??

Because both these cards are setup for 8bit rgb, yet the ATI results in the +10% , quite large a swing.

[huhn]

nvidia has full support for the HDMI content type so try that. afaik AMD doesn't have an option for that maybe in registry like so many more options.

[tp4tissue]

Originally Posted by huhn

nvidia has full support for the HDMI content type so try that. afaik AMD doesn't have an option for that maybe in registry like so many more options.

What do you mean by hdmi content type.

I've set both outputs to RGB 32bit 8 bit full based on your previous recommendation ?

[Warner306]

Originally Posted by tp4tissue

Yea, that's what I meant, create an HDR 3Dlut, but ONTOP of Dynamic Tonemapping, output HDR.

Do you know if this feature will be added in the future ?

Seems to be straightforward, is there a specific hitch that it's not already possible ?

I was going to answer your question earlier but missed it.

It is hard to use a 3D LUT this way because DisplayCAL enforces its own roll-off. I don't know how you would do it without excessive doubling processing. The 3D LUT could maybe correct the white balance and white point, but not alter the transfer function with its own roll-off.

[tp4tissue]

Originally Posted by Warner306

I was going to answer your question earlier but missed it.

It is hard to use a 3D LUT this way because DisplayCAL enforces its own roll-off. I don't know how you would do it without excessive doubling processing. The 3D LUT could maybe correct the white balance and white point, but not alter the transfer function with its own roll-off.

Why does the tone map function only work with an SDR lut ?

For example, on many tvs, in HDR input mode, the Gamut is larger than its SDR mode.

So, after madvr does the tone map, instead of going to an sdr calibrated table before output, it will go to a wider hdr calibrated table.

[el Filou]

Originally Posted by tp4tissue

What do you mean by hdmi content type.

It's an option in HDMI where the source can specify what type of content is sent, e.g. computer graphics/movie/photos/auto.
With NVIDIA you can change that setting in the control panel. Most TVs use a different kind if picture processing depending when that setting is on something other than auto.

[huhn]

how do you stop a TV from dynamic tone mapping in HDR mode.

if you do dynamic tone mapping and send this technically wrong image to the TV the TV will do again dynamic tone mapping making it even more wrong so waht'S the point of an 3D LUT here.

just set your Tv up for native gamut or send bt 2020 so it will use the full gamut.

who even told you there is an wider "calibration table" for HDR?

What do you mean by hdmi content type.

http://abload.de/img/hdmi-content-typebtumj.png

first result in an search engine

[huhn]

Originally Posted by el Filou

It's an option in HDMI where the source can specify what type of content is sent, e.g. computer graphics/movie/photos/auto.
With NVIDIA you can change that setting in the control panel. Most TVs use a different kind if picture processing depending when that setting is on something other than auto.

just so you know i have a PC monitor that is resyncing when it is on auto everytime a video is played and stopped.

so auto means nvidia or other software is able to actively change that.

[tp4tissue]

Originally Posted by huhn

who even told you there is an wider "calibration table" for HDR?


http://abload.de/img/hdmi-content-typebtumj.png

first result in an search engine

It's not a wider table, but the gamut is wider and different between SDR and HDR modes.

Even in the various different sdr modes, tvs can have different gamuts.



If I set the HDMI content type on AUTO, what is it defaulting to when running madvr (Movie ? or Desktop Program)

[huhn]

no clue.

edit: madVR FSE is triggering "games" with FSE and i can't find anything with windowed mode.
is so advance that leaving it will disabled gaming mode in my TV even if i manually enabled it before.
setting the output type to movies in the GPU driver switches my TV into cinema preset(d80 whitepoint close panasonic close call...) and my TV even shows an notification. i didn't find a program that does trigger this in the GPU driver.

so madVR may trigger gaming mode in the nvidia driver which makes sense if you think about it.

[chros]

Originally Posted by huhn

edit: madVR FSE is triggering "games" with FSE and i can't find anything with windowed mode.
is so advance that leaving it will disabled gaming mode in my TV even if i manually enabled it before.
setting the output type to movies in the GPU driver switches my TV into cinema preset(d80 whitepoint close panasonic close call...) and my TV even shows an notification. i didn't find a program that does trigger this in the GPU driver.

so madVR may trigger gaming mode in the nvidia driver which makes sense if you think about it.

Interesting finding, thanks, I'll try it out tonight with my old LG TV, maybe ...

[Alexkral]

Originally Posted by Warner306

Soulnight's tool increases or decreases the target nits. Increasing the target nits raises the knee point further up the PQ curve to reduce compression, but it also changes the absolute brightness of all values relative to the output display.

Actually what you are changing with the target peak nits parameter is the absolute (only relative to the mastering display luminances) value of the knee point, which affects all the values above it. After that, all the image values need to be normalized to the [0, 1] SDR relative range, which raises the global brightness. That's why I thought that the dynamic curves where doing the same, but after reading in more detail the specs this doesn't seem to be the case. For Dolby Vision the target luminances don't change, but for HDR10+ there is indeed a parameter called TargetedSystemDisplayActualPeakLuminance which is a 2D LUT that affects all values, and is intented to take into account the "peak luminance that a display is capable of delivering while rendering the scene", which "depends on the spatial distribution of the luminance levels of the pixels in the scene and the power consumption limits of the display".