madVR - high quality video renderer (GPU assisted)

[thighhighs]

Originally Posted by Sunspark

This forest picture is ridiculous. Even at 400% magnification, the only change I noticed was the black on the rim around the wizard's hat changing in thickness which could just simply be a rendering artifact and maybe, an incredibly subtle fraction of a fraction change in texture of stuff. The fact that I say maybe is how indistinguishable it is.

Don't shave hairs this finely. The power consumption isn't worth it at all especially if you can't see any changes.

Just to clarify: and nothing else at all? The fact that there are 3 different algorithms does not bother you?

Originally Posted by ryrynz

I don't downscale any more as I have a 4K set, but 2D was quite superior to 1D. After my initial review of AB I decided I didn't find it useful, better imo to just reduce the sharpness of the display, but it all depends.

AB changes the result of the algorithm.
For the screen I using the "source as is" setting. But it's not 0. If it does anything, I don't expect much. To "fix" the 2D100, I need to gimped the screen so that I can't see anything at all And even then I could see AB/no AB.
Why configure the screen based to SSIM2D at all? If the special patterns are ok, everything is fine.

[ryrynz]

Stick with whatever works for you, your content and display, that's all that matters.. The differences are close to insignificant.

[Magik Mark]

Hi!

Would u consider rtx 3000 series with or without LHR? Does it make any difference?

[thighhighs]

Originally Posted by ryrynz

Stick with whatever works for you, your content and display, that's all that matters.. The differences are close to insignificant.

But that's not what you meant before. You said (again) the SSIM2D can't look weird, it's just that my screen isn't set up properly or something.
I understand that you liked it and everything looked great for you. So AB doesn't make sense. I can accept it.
This is the only algorithm in madVR with this level of sharpness. This is insane changes. But if you don't see sharpening artifacts, you don't think the result looks weird. There is something similar with NGU Sharp. As you know NGU Sharp has an cartoon-look problem. Does it has sharpening artifacts or does it just look weird? Is this really a problem? Do all users notice this when they watch their videos? Or does it just look great to them? That's all I wanted to say.

[flossy_cake]

What about MadVR's suite of sharpening enhancements in processing -> image enhancements? Do you not like it? I find it gives me more control - I can choose the exact type and the amount of sharpening I want. Has saved many DVDs from looking soft and blurry.

[thighhighs]

flossy_cake
Personaly I do not use additional processing (sharpening) at all. I use NGU AA for SD and Standard for 720p. And very satisfied.
For 1080 -> 2160 I would like to use NGU Sharp. I did this for the 2160 screen. But my screen is now 1080.

For 2160 -> 1080, a sharp algorithm is also enough for me. This is how I watch my favorite movies. madVR gives me an advantage over 1080 blu-ray. For example, Dune 4K old/new, Jaws 4K, Predator 4K, Alien 4K, the Thing 4K, The Hobbit 4K and others. Also Doom 4K and RE2 4K are actualy "upscaled-4k" but look much better.
The only time I'm really disappointed is Monster Hunter 4K. The result is too dim even with 100nits. 1080 blu-ray is much better, but maybe it's not madVR's fault, I don't know.

[ashlar42]

Originally Posted by thighhighs

flossy_cake
Personaly I do not use additional processing (sharpening) at all. I use NGU AA for SD and Standard for 720p. And very satisfied.
For 1080 -> 2160 I would like to use NGU Sharp. I did this for the 2160 screen. But my screen is now 1080.

For 2160 -> 1080, a sharp algorithm is also enough for me. This is how I watch my favorite movies. madVR gives me an advantage over 1080 blu-ray. For example, Dune 4K old/new, Jaws 4K, Predator 4K, Alien 4K, the Thing 4K, The Hobbit 4K and others. Also Doom 4K and RE2 4K are actualy "upscaled-4k" but look much better.
The only time I'm really disappointed is Monster Hunter 4K. The result is too dim even with 100nits. 1080 blu-ray is much better, but maybe it's not madVR's fault, I don't know.

I'm no expert but if you choose NGU Sharp over Soft, Standard or Anti-Alias it's pretty clear you, in fact, are applying some form of sharpening to the picture. Just using a different route. And there's where flossy cake suggestion comes into its own. You might have overlooked options that allow you a finer control on what happens to the picture.

[flossy_cake]

These are my sharpening presets in case anyone finds them useful. I tuned them so that for example the mid setting would appear subjectively similar "strength" on SD content regardless of algorithm.

Lumasharpen low: 0.65
Lumasharpen mid: 1.30
Lumasharpen high: 1.95
Lumasharpen max: 2.60

Adaptivesharpen low: 0.2
Adaptivesharpen mid: 0.4
Adaptivesharpen high: 0.6
Adaptivesharpen max: 0.8

Customsharpen low (sharpen, crispen, thin, enhance): 0.5, 0.5, 0.5, 0.5
Customsharpen mid (sharpen, crispen, thin, enhance): 1.0, 1.0, 1.0, 0.5
Customsharpen high (sharpen, crispen, thin, enhance): 1.5, 1.5, 1.0, 0.5
Customsharpen max (sharpen, crispen, thin, enhance): 2.0, 2.0, 1.0, 0.5

Antibloat filter: unticked for all
Antiringing filter: ticked for all

I subjectively prefer Customsharpen low and mid for SD content. The "thin edges" seems good at creating the illusion of higher resolution image, but beware it does change the appearance of fonts quite a bit which is why I cap it at 1.0. I'm also capping "enhance detail" at 0.5 as it enhances compression artefacts, but it is good at creating illusion of detail in textures like wood grain or stone texture in walls, so if you have a clean source you can run it higher.

As I said previously I generally don't feel the need to use sharpening for native HD sources so I'm not a sharpening whore or anything. I just subjectively feel SD on a big screen is problematic and needs something extra. A lot of the time the sharpening will be baked into the source by the company that did the mastering and I don't feel those particular shows or movies need anything extra. eg. was watching I'm Alan Partridge S01E01 yesterday and didn't feel it needed any extra sharpening, seems to have sharpening baked into the source already.

[sebna]

When I tried to experiment with downscaling 4k to 1080p (my PJ is native 1080p) I found that limited options for downscaling in compare to upscaling were not producing natural looking SQ at all. I found BD rips played native 1080p better looking than 4k rips downscaled to 1080p.

Was I doing it wrong and it is matter of finding sweet spot of settings or is it expected for it to be as described? Any good settings recommendation for downscaling as described?

Also is 4070ti performance with MadVR already known? Would you choose it over 3090 / ti?

Thanks

[sebna]

Originally Posted by Sunspark

kopija: You should see if your monitor can do 23.976/24/25 refresh rates. Mine is able to on the HDMI input. If it is able to, then you never need to use smooth motion, etc.

But how do you force PC to output 23.976hz? Or are you happy with 23.976fps to be sent as 24hz refresh rate?

[flossy_cake]

Originally Posted by sebna

When I tried to experiment with downscaling 4k to 1080p (my PJ is native 1080p) I found that limited options for downscaling in compare to upscaling were not producing natural looking SQ at all. I found BD rips played native 1080p better looking than 4k rips downscaled to 1080p.

Was I doing it wrong and it is matter of finding sweet spot of settings or is it expected for it to be as described? Any good settings recommendation for downscaling as described?

I'm struggling to think of a reason why downscaling 4k to 1080p would cause issues, in fact it should look even better than native 1080 given the supersampling effect. I've previewed some 4k movies downscaled to 1080 using Lanczos and the result looks very good, but all of the algorithms look good and I don't think you should need to tweak anything. What do you mean by "not natural"? Can you post a screenshot?

[Sunspark]

Originally Posted by sebna

Any good settings recommendation for downscaling as described?

Bicubic50/Catmull-Rom, scale in linear light checked, activate anti-ringing filter relaxed checked.

Originally Posted by sebna

But how do you force PC to output 23.976hz? Or are you happy with 23.976fps to be sent as 24hz refresh rate?

With a hdmi connection to the display, if you do the switching in madvr you would find that in the devices display modes section of madvr. For 23.976 you need to write 1080p23

You can also set it up in the player instead, playback fullscreen section.

Your milage may vary as it depends to a certain extent on what edid blocks are present. This won't work with dvi for example, with that you'd have to do futzing about with the video driver instead defining a custom resolution and even then it might not let you.. but happily, on my display it already had support for it on the hdmi connection.

[sebna]

Originally Posted by flossy_cake

I'm struggling to think of a reason why downscaling 4k to 1080p would cause issues, in fact it should look even better than native 1080 given the supersampling effect. I've previewed some 4k movies downscaled to 1080 using Lanczos and the result looks very good, but all of the algorithms look good and I don't think you should need to tweak anything. What do you mean by "not natural"? Can you post a screenshot?

Thanks I have to do more testing. I think I had issues with it looking like painted with thick brush strokes. It looked fat and unnatural. I have to try different algorithms.

Am I right to assume that it is possible to use tone mapping with downscaling at the same time?

[sebna]

Originally Posted by Sunspark

Bicubic50/Catmull-Rom, scale in linear light checked, activate anti-ringing filter relaxed checked.



With a hdmi connection to the display, if you do the switching in madvr you would find that in the devices display modes section of madvr. For 23.976 you need to write 1080p23

You can also set it up in the player instead, playback fullscreen section.

Your milage may vary as it depends to a certain extent on what edid blocks are present. This won't work with dvi for example, with that you'd have to do futzing about with the video driver instead defining a custom resolution and even then it might not let you.. but happily, on my display it already had support for it on the hdmi connection.

Thanks, I will try those down scaling algorithms. I assume I can still use tone mapping while downscaling?

Would 1080p24 be not more correct with 23.976 being more 24p than 23p?

Thanks

[brazen1]

Originally Posted by sebna

Thanks, I will try those down scaling algorithms. I assume I can still use tone mapping while downscaling?

Would 1080p24 be not more correct with 23.976 being more 24p than 23p?

Thanks

No. There is 23.976 video and 24.000 video.

[flossy_cake]

23.976 and 24.000 are two completely different formats that aren't compatible with eachother. If you play back 23.976 at 24.000hz you will have 1 repeated frame every 41.7083 seconds. The vast majority of movies and shows are 23.976 (specifically 24.000/1.001, and yes even this precision matters otherwise if you are off by 0.001hz you will still have a frame skip every 16 minutes and if it happens when the camera happens to be panning it is noticeable). Then there are some European releases of Blurays like eg. those from Studio Canal are 24.000 so you'll need to setup 2 display modes to get them both playing perfectly & this is what I did using Custom Resolution Utility but it is very difficult and needed a very specific settings combination to work and what works for one system won't necessarily work for another.

edit: forgot to mention you can use an Avisynth script to speed up or slow down the frame rate (the audio will get sped up and slowed down to remain in sync) eg. this is for slowing down 25.000 to 23.976:

DirectShowSource("C:\Some 25fps video.mkv")
AssumeFPS(24000, 1001, true)
# http://avisynth.nl/index.php/FPS#AssumeFPS

Then you just save the above as a .avs file and open it with the media player.

[Asmodian]

Originally Posted by sebna

Would 1080p24 be not more correct with 23.976 being more 24p than 23p?

23 Hz in Windows means 23.976 Hz, not 23.000 Hz.

Or at least close, since the GPU never does a perfect refresh rate.

[sebna]

Originally Posted by flossy_cake

23.976 and 24.000 are two completely different formats that aren't compatible with eachother. If you play back 23.976 at 24.000hz you will have 1 repeated frame every 41.7083 seconds. The vast majority of movies and shows are 23.976 (specifically 24.000/1.001, and yes even this precision matters otherwise if you are off by 0.001hz you will still have a frame skip every 16 minutes and if it happens when the camera happens to be panning it is noticeable). Then there are some European releases of Blurays like eg. those from Studio Canal are 24.000 so you'll need to setup 2 display modes to get them both playing perfectly & this is what I did using Custom Resolution Utility but it is very difficult and needed a very specific settings combination to work and what works for one system won't necessarily work for another.

edit: forgot to mention you can use an Avisynth script to speed up or slow down the frame rate (the audio will get sped up and slowed down to remain in sync) eg. this is for slowing down 25.000 to 23.976:

DirectShowSource("C:\Some 25fps video.mkv")
AssumeFPS(24000, 1001, true)
# http://avisynth.nl/index.php/FPS#AssumeFPS

Then you just save the above as a .avs file and open it with the media player.

Very interesting so http://avisynth.nl/index.php/FPS#AssumeFPS does what ReClock used to do? Or does it work differently?

Also extremely interesting if true what you guys are saying about refresh rates.

Let me rehash it - you claim that 23hz setting in NV CP is not 23hz but in fact 23.976? If so why is not called so?

I am asking because I always though (based of name of the option) that 23hz in CP is 23hz refresh rate is what FPS is sent to the display device which gets a flag for 23hz.

But you say that NV is actually sending cinematic 23.976 FPS? Are you sure?

If I am right than less problems would cause 24hz setting as obviously it is much closer to 23.976 than 23hz.

Thanks

[Klaus1189]

Originally Posted by sebna

Also extremely interesting if true what you guys are saying about refresh rates.

That is true
Just look up advanced display settings of Windows 10. I made a screenshot of mine:
https://drive.google.com/file/d/1lq2...usp=share_link

 23 Hz ->  23.976 Hz -> exactly 24/1.001 Hz
 24 Hz ->  24.000 Hz -> exactly 24 Hz
 25 Hz ->  25.000 Hz -> exactly 25 Hz
 29 Hz ->  29.970 Hz -> exactly 30/1.001 Hz
 30 Hz ->  30.000 Hz -> exactly 30 Hz
 50 Hz ->  50.000 Hz -> exactly 50 Hz
 59 Hz ->  59.940 Hz -> exactly 60/1.001 Hz
 60 Hz ->  60.000 Hz -> exactly 60 Hz
100 Hz -> 100.000 Hz -> exactly 100 Hz
119 Hz -> 119.880 Hz -> exactly 120/1.001 Hz
120 Hz -> 120.000 Hz -> exactly 120 Hz

Kodi has also a bug in it, but it seems that nobody cares:
https://forum.kodi.tv/showthread.php?tid=370505

Just like for frame adaptive deinterlacing - Nobody cares
https://forum.kodi.tv/showthread.php?tid=370507
I am so glad to be here

Originally Posted by sebna

Let me rehash it - you claim that 23hz setting in NV CP is not 23hz but in fact 23.976? If so why is not called so?

23 Hz is used to let you know that it is 23.976 and not 24.000p.
24 Hz is for 24.000p
23.976 is exactly 24/1.001
just like 29 Hz is 29.970 and exactly 30/1.001
just like 59 Hz is 59.940 and exactly 60/1.001
just like 119 Hz is 119,880 and exactly 120/1.001

Originally Posted by sebna

I am asking because I always though (based of name of the option) that 23hz in CP is 23hz refresh rate is what FPS is sent to the display device which gets a flag for 23hz.

But you say that NV is actually sending cinematic 23.976 FPS? Are you sure?

If I am right than less problems would cause 24hz setting as obviously it is much closer to 23.976 than 23hz.

Have you ever looks at OSD stats in madVR? press CTRL + J during playback and take a look for yourself

[flossy_cake]

Originally Posted by sebna

Very interesting so http://avisynth.nl/index.php/FPS#AssumeFPS does what ReClock used to do? Or does it work differently?

I don't know how ReClock works but it's 32-bit only (not compatible with 64-bit media players) so that's why I never used it. But I'd prefer to use Avisynth anyway as I feel like I'm more in control. You can also do pitch correction with eg. TimeStretch(pitch=((25000.0/1000.0)/(24000.0/1001.0))*100) which increases the pitch from 23.976 to 25.000 without changing the length of the audio, which is useful for example if a 25.000 video already had pitch correction baked into it by the company that did the 23.976 -> 25.000 conversion in the first place (otherwise the pitch would be too low when you slowed it back down to 23.976 with AssumeFPS(24000, 1001, true)).

Originally Posted by sebna

But you say that NV is actually sending cinematic 23.976 FPS? Are you sure?

In addition to ctrl+J you can also measure it at www.vsynctester.com