madVR Options Explained

[SweetLow]

"enable smooth motion frame rate conversation only if there would be motion judder without it or if the display refresh rate is an exact multiple of the movie frame rate."

I understand this two options exactly the same.
But what about "always"?
Now i suppose - in this mode madVR ignore source rate (and simply construct frames with target rate). So we never get dropped/repeated frame.
Is this true?

[huhn]

that's true.

[Asmodian]

Updated formatting and some descriptions and added that black bar detection doesn't work with DXVA Native.

[huhn]

shifting the image does works with DXVA so hmm.

[Asmodian]

I will do some testing and update it to be more specific, thanks.

Edit: Which image shifting did you test? I cannot get anything in "automatically detect hard coded black bars" to function when using DXVA2 Native, the always shift the image option doesn't have an effect. Even the hidden option ShowBlackBars doesn't do anything when using DXVA2 Native, it looks like detect hard coded black bars doesn't run at all when using DXVA2 Native.

[huhn]

a screen should be better:
https://abload.de/img/imageshiftqcs9f.jpg

and yes hard coded bars are not detected.

"cleanup image boarders by..." and "if there are big black bars:" works with DXVA native too. i even forgot they exist.

maybe something like: most black bar detection features doesn't work with DXVA native(DXVA copy back is recommended) in a proper english is a enough or it is getting complicated in the end.

[SweetLow]

shifting the image does works with DXVA so hmm.
Of course. Basic logic tell us - this option have nothing related to black bars _detection_ and fully independent of this.
And I don't understand, why madshi put this option in group of "detect hardcoded bars" (and disable image shifting when detect option disabled). May be ask him in main thread?

[huhn]

they where added at about the same time and he said before that he wants to rework/rearrange all these option/menus at one point and i'm pretty sure he looks over this in thread from time to time anyway.

but the place is not that far of. remove the words "hard coded" and it makes total sense but a lot of thy other option could be confusion. in the end this is about black bars.

madshi had problem with place on settings pages too and the first thing that comes up to my mind was "just add tabs" but in the end i'm pretty sure creating these menus is not fun at all and creating new algorithms feels more rewarding and make more fun.

i will come up on these topic when he is actively working on these options.

[SweetLow]

i will come up on these topic when he is actively working on these options.
Ok. But i prefer change this in one simple combobox: "Image position on Screen" with 3 values "Top","Center" and "Bottom"

[huhn]

if hard coded black bars are not the problem them the player should do that in my eyes.

i'm pretty sure mpc-hc can only do this manually maybe another play can do that.

what stops you from using this current option?

off is centered.

[Asmodian]

a screen should be better:
https://abload.de/img/imageshiftqcs9f.jpg

and yes hard coded bars are not detected.

"cleanup image boarders by..." and "if there are big black bars:" works with DXVA native too. i even forgot they exist.

maybe something like: most black bar detection features doesn't work with DXVA native(DXVA copy back is recommended) in a proper english is a enough or it is getting complicated in the end.

Ah, right, it only applies to the black bars applied by madVR. That gets complicated. Cleanup seems worthing mentioning too.. hmm..

[Asmodian]

Updated for madVR v0.91.0. Next Generation Upscaling (NGU)!

[Asmodian]

Updated for madVR v0.91.4. Big changes to the image doubling settings, new "let madVR decide" settings which I do recommend and use myself.

edit: I forgot to upload my new settings.bin, now corrected. :o

[Hrxn]

BTW, what happens if you use such a settings.bin? Usually the preferences are stored in the registry, do they get overwritten?
Or works it like this: If settings.bin present, load settings stored there, if not found, use registry values?

[sneaker_ger]

IIRC: both have a value to store how recent they are. The most recent one overwrites the other (or create the other if does not exist).

[Asmodian]

Yes, for this reason resetting madVR to defaults before copying in a settings.bin is recommended.

[Kitty]

As explained "Image Upscaling" is how madVR scales to the target resolution if it is larger than the source. So my question is, what's the point of "Chroma Upscaling"? It's explained as, how madVR doubles the chroma resolution in both dimensions. There's no need for it if "Image Upscaling" does the same. So does it serve an alternate purpose, it'd be greatly appreciated if someone would explain.

[Asmodian]

Chroma is stored at half the resolution of luma in consumer formats (called 4:2:0 sampling, found on DVD, blurry, streaming, etc.) so you first need to scale the chroma to match the luma before the conversion to RGB. You often want to use a different method for this compared to scaling both the chroma and luma together, as is done by image upscaling.

Referring to the image in the third post might help you understand the steps needed. :)

[Kitty]

Chroma is stored at half the resolution of luma in consumer formats (called 4:2:0 sampling, found on DVD, blurry, streaming, etc.) so you first need to scale the chroma to match the luma before the conversion to RGB. You often want to use a different method for this compared to scaling both the chroma and luma together, as is done by image upscaling.

Referring to the image in the third post might help you understand the steps needed. :)

I see, that's very interesting, I didn't know that. Thank you very much for the info. :)

[Asmodian]

New flowchart, people were often confused by the old one so maybe this one will be better understood. I think I was trying to pack too much into it, now it is really only an example using NGU for image upscaling.

Old:
http://i1222.photobucket.com/albums/dd496/asmodian3/madVR%20Chart_zpsj6tljmro.png

New:
http://www.mediafire.com/file/kp43u45i8a3ayqt/madVR_Chart_v2.png

[Kitty]

I seemed to have forgot to ask about this before, but I'd like to know the proper procedure for the "Scaling Algorithms." I use NGU and I was wondering if I should max out the "Chroma Upscaling" all the way before touching the "Image Upscaling" or should I move them together. Basically what I'm asking, is if the "Chroma Upscaling" more important to image quality than "Image Upscaling."

So I have a couple examples and I'd like to know which is optimal:

"Chroma Upscaling" = "NGU - very high"
"Image Upscaling" = "NGU - Algorithm Quality -> low"

"Chroma Upscaling" = "NGU - medium"
"Image Upscaling" = "NGU - Algorithm Quality -> high"

Which option is the better option of the two?

[huhn]

chroma is very unimportant.

you should always judge the picture quality by your self.
increasing the scaling algorithm until the GPU is at his max is not a proper way to get a good result.

NGU for example is not good for all types of sources.

[Kitty]

chroma is very unimportant.

you should always judge the picture quality by your self.
increasing the scaling algorithm until the GPU is at his max is not a proper way to get a good result.

NGU for example is not good for all types of sources.

Don't get me wrong, I understand that luma is much more important than chroma, however I just thought that since we had to upscale chroma to RGB first that it might hold some importance. Also, I'm probably not maxing out my GPU in the sense that you are thinking. I press "CTRL + J" to get the OSD and I look at the frame time to make sure I'm a couple of milliseconds lower.

I primarily watch animated content, such as anime, and NGU in my testing is perfectly fine with it. However, the reason I asked about what was better between my options, which you didn't answer btw :P, was because in my testing I noticed little to no difference between "Chroma Upscaling - low" and "Chroma Upscaling - very high" even when looking an inch away from my monitor and was just wondering theoretically if it mattered more.

After some current testing I think you're right in saying that chroma is less important. I kept the "Chroma Upscaling" at "very high" and just modified the "Image Upscaling" algorithms option. I noticed more of a difference between the options than when I was modifying the "Chroma Upscaling."

So in conclusion, I'm pretty sure that the second set of settings that I proposed in my original question would be the better choice. Producing more noticeable practical results. If anyone else wants to add anything, if I might be missing something, then go right ahead. :)

[Asmodian]

You are correct, go with your second option. Chroma Upscaling is more important than the chroma algorithm quality in image upscaling but it is still not something to spend too much GPU power on.

I like NGU med or high for chroma upscaling but only if the luma quality in image upscaling is already on very high. :)

NGU is great for HD anime but it is too sharp for SD anime, IMO, it shows all the artifacts and oddities too well. NNEDI3 does better with SD anime.

[Asmodian]

Added some comments on the "trade quality for performance options" (https://forum.doom9.org/showthread.php?p=1709814#post1709814).

[Asmodian]

Updated for madVR v0.91.5.

NGU can do x4 in one step now (when chroma quality = "normal"). It isn't as sharp and exaggerates artifacts much less than two x2 steps. :)

No other changes.

[Warner306]

New flowchart, people were often confused by the old one so maybe this one will be better understood. I think I was trying to pack too much into it, now it is really only an example using NGU for image upscaling.

Old:
http://i1222.photobucket.com/albums/dd496/asmodian3/madVR%20Chart_zpsj6tljmro.png

New:
http://www.mediafire.com/file/kp43u45i8a3ayqt/madVR_Chart_v2.png

According to this, chroma is downscaled after conversion to RGB. If that is true, why doesn't the madVR OSD report this?

http://i1357.photobucket.com/albums/q760/Warner306/madVR%20Guide/NGU-image--doubling_zpsitugr0yk.png

Shouldn't it be Bicubic60 AR < SSIM1D100 AR?

I admit, I don't know this at this stage as a user.

[huhn]

chroma and luma are totally different things.
and chroma is not downscaled only luma at one point to match the chroma/target resolution.

[Asmodian]

According to this, chroma is downscaled after conversion to RGB. If that is true, why doesn't the madVR OSD report this?

http://i1357.photobucket.com/albums/q760/Warner306/madVR%20Guide/NGU-image--doubling_zpsitugr0yk.png

Shouldn't it be Bicubic60 AR < SSIM1D100 AR?

I admit, I don't know this at this stage as a user.

Actually it can work both ways, depending on the trade quality for performance option "scale chroma separately, if it saves performance". My chart shows as it would work with this disabled. It does default to enabled.

The OSD doesn't actually ever show downscaling for chroma. Using rendering times and a heavy chroma doubler it is easy to see chroma doubling or not when toggling the trade quality for performance option but the OSD chroma line doesn't change. If both luma and chroma are set to the same doubler the OSD matches my chart and you can see chroma doubling activate and deactivate when toggling "scale chroma separately, if it saves performance"; the OSD line changes from "image" to "luma" when enabled but it still doesn't indicate what happened to the chroma (upscaled, downscaled, or how).

"scale chroma separately, if it saves performance" Disabled:
https://www.mediafire.com/?6rzcyppdks1tk22
"scale chroma separately, if it saves performance" Enabled:
http://www.mediafire.com/?u5hz2usl4hkkx99

I wanted to show chroma doubling enabled to indicate where the performance hit comes from and this is more representative of how it works when doubling without downscaling a lot afterwards. It is hard to come up with a good chart that explains enough without being hard to understand. :o

[mboxer511]

Great post. This was really helpful as was your guide you posted on wiki.jriver.com which led me to this forum. There were a couple questions/corrections/guide mistakes I wanted to communicate with you:

Smooth Motion -

I believe you might have made a mistake/typo in your description about smooth motion in this line:

"Note that smooth motion has less blurring when displaying 23.976 fps at 60 Hz than 23.976 fps at 23.970 Hz so this option..."

I think you meant 29 fps at 60 Hz (or something to that effect).

A suggestion I had (because this really helped me when I read this in you guide) is related to:

enable automatic full screen exclusive mode -

In your guide you suggest turning this off which is actually very helpful as it allowed me to use the full media bar in full screen when hover over it with a mouse. It also removes any delay between switching between full and windowed. Perhaps you might want to mention that to some effect in your guide under the description of that mode.

Finally I had a question and was hoping you could explain what exactly "activate 200% super sampling" in image upscaling does? I have a decent rig (Nvidea GTX 960 GPU, i5 6600k OC @ 4.6Ghz, 16GB of RAM OC @ 3000MHz on a MSI Z170 M5 Motherboard) and yet saw a severe drop in performance when enabling that feature with no quality improvement. It was actually one of the rare times any setting change effected my video playback performance.

[Asmodian]

Great post. This was really helpful as was your guide you posted on wiki.jriver.com which led me to this forum. There were a couple questions/corrections/guide mistakes I wanted to communicate with you:

I am glad it helped.. but I don't remember posting anything on that wiki. I am glad whoever did thought it was valuable too. ;)

Smooth Motion -

I believe you might have made a mistake/typo in your description about smooth motion in this line:

I think you meant 29 fps at 60 Hz (or something to that effect).


I didn't make a typo actually, the use of the same source frame rate for both examples was very deliberate. You don't want to use smooth motion when the source frame rate almost equals the display's refresh rate but it is great when the display's refresh rate is much higher. :)

edit: That smooth motion works better for 24p into 60p than 23.976p into 24p is somewhat counter intuitive, and was the point of that example. I will change the rates to these simpler ones make this point better, even if 24p video is quite rare. ;)
edit2: Hmm, I don't like using different source frame rates in that example for some reason, hopefully just using 24 Hz will help.


A suggestion I had (because this really helped me when I read this in you guide) is related to:

enable automatic full screen exclusive mode -

In your guide you suggest turning this off which is actually very helpful as it allowed me to use the full media bar in full screen when hover over it with a mouse. It also removes any delay between switching between full and windowed. Perhaps you might want to mention that to some effect in your guide under the description of that mode.

I do suggest turning off exclusive mode, and I don't use it myself. I updated the description to stress the downsides a bit more. These posts aren't meant to be a guide but I do want people to understand the consequences of each option.


Finally I had a question and was hoping you could explain what exactly "activate 200% super sampling" in image upscaling does? I have a decent rig (Nvidea GTX 960 GPU, i5 6600k OC @ 4.6Ghz, 16GB of RAM OC @ 3000MHz on a MSI Z170 M5 Motherboard) and yet saw a severe drop in performance when enabling that feature with no quality improvement. It was actually one of the rare times any setting change effected my video playback performance.

"activate 200% super sampling" doubles the image all the time, even if it doesn't need upscaling at all. Normally doubling will only trigger if it needs upscaling 1.2x or more.

For 1080p video even on a 1080p display this means that the video will be upscaled to 4K with the doubling algorithm, any image enhancements applied, and then downscaled to 1080. My Titan XP can barely do that with NGU very high.

Personally I don't find 200% supersampling very useful, it certainly isn't useful all the time. The only situation I would think might be useful was blurry but high res video you wanted to sharpen a lot. I tend to prefer sharpening done at higher resolutions and then downscaling but not everyone agrees with that.

edit: Thanks! I missed a description for 200% super sampling. :o

[cyber201]

More Advanced Topics

My settings.bin (https://www.mediafire.com/?2c118c36s25w554) (v0.91.5), optimized for the Nvidia Titan X (Pascal), Windows 10, and up to a 4K display. Profiles for SD, 720, 1080, UHD, and above. Simply restore madVR to default settings and place it in the madVR program directory. madVR should also have write access to its directory (e.g. not in "program files"). :)


Hi, thanks for the config file.
My setup is Windows 10 + GTX1070 + i5 KabyLake + 16Gb DDR4 + TV UHD Samsung KS8000 + Denon X3300W.

I have to do a modification of the upscaling of 4K and 1080 because the video card reach 70°c and the GPU fans make too much noise.

Can you give me your support to configure at the best the settings for the 1080p and 4k profiles?

Thanks a lot

[Asmodian]

If temperature is a problem you might want to simply use the defaults, the high quality algorithms in madVR do use a lot of power. My Titan XP also runs hot with madVR, though it is water cooled so never goes above 40C.

That said, I would simply turn down the NGU quality for both chroma and image upscaling to reduce the power requirements. Also change the dithering to ordered for both profiles.

Edit: I should have mentioned that Jinc is a very nice all purpose algorithm. It is a great option if you want to keep temps down with a high end GPU while offering a significant improvement over the defaults.

[Asmodian]

Finally updated for madVR v0.91.7, sorry for the delay.

NGU Anti-aliasing, NGU Sharp quadrupling for all quality levels, and a new image upscaling settings page. :)

I have also updated my settings.bin file.

[cyber201]

Thanks a lot....
You are great

[Kitty]

I'm using MPC-BE, however I was curious what the "Vsync" option does and how it ties in with MadVR? Does it have any affect, since the player is using MadVR instead of it's internal renderer?

Also, if I'm using ReClock, how does it's "Vsync Correction" affect MadVR?

[Asmodian]

Sorry, I do not use MPC-BE at all and do not know how it's Vsync options interact with madVR.

I also don't use ReClock anymore, madVR's smooth motion is a better option for handling a mismatched Vsync. madVR does not seem to notice ReClock's Vsync correction so do not use it at the same time as smooth motion.

[Kitty]

Sorry, I do not use MPC-BE at all and do not know how it's Vsync options interact with madVR.
Well, MPC-BE is just a fork of MPC-HC, but I don't know if you use MPC-HC either.

By the way, what media player are you using, and if applicable, are you using a separate audio renderer or just the internal one?

[Asmodian]

I use Zoom Player Max and its internal audio renderer.

[Asmodian]

Updated for madVR v0.91.8. NGU Soft and Standard got implemented. :D
Also improved quality for NGU Sharp 4x quadrupling at very high.

[cyber201]

Yeahhh
Thanks a lot

[cyber201]

Hi Asmodian, thanks for your work.
One question. How can I apply the best setup for this two parameters?

CPU queue size: [16] The buffer in system memory for decoded video and xySubFilter subtitles.
GPU queue size: [8] The buffer in the graphics memory for madVR’s rendering.

I have a i5-7600K CPU and a GTX 1070 for GPU

With your setitings, sometimes I have 5/10 dropped or represented frames. Why?
I have setup the medium quality for chroma ad upscaling for 1080 profile and low quality for 4K. But the problem of the dropped frames is the same.

Thanks a lot for your support.

[Asmodian]

Normally some frames are dropped or have presentation glitches during seaking and startup, is this what you are seeing or do you get them during normal playback?

Queue sizes are a trade-off between responsiveness and stability, I tend to go for smaller queues like 12 CPU, 6 GPU and only 3 frames presented in advance for max responsiveness. My system is still very stable during playback, with no dropped frames or presentation glitches, but I do get a few frame drops and glitches on startup or when seaking.

I have noticed that Windowed Overlay is a little more stable than other modes, remember exclusive overwrites overlay.

[cyber201]

I have this problem during normal playback.
When i found this option? Windows overlay?
Madvr is setup for exclusive mode

Thanks

[Asmodian]

Window overlay is on the same settings page as exclusive mode.

Do you have your GPU's power setting on Adaptive? Power optimized often clocks the GPU too low.

If you are using my settings file you probably need to lower some settings, the 1070 is a lot slower then the Titan.

[cyber201]

Yes, I use your config file with a little downgrade in quality for the profile 1080 and 2160 due to my GTX1070.
My GPU power is set to Adaptive and now I have exclusive mode enable. For you i need to try overlay?

Thanks a lot

[Asmodian]

Yes, try overlay if you are still having issues. You need to turn off exclusive and DX11 for overlay to work.

[cyber201]

If i enable only windows overlay, the screen is all green.
Now I have enable exclusive and dx11.
With this settings, the hdr metadata and the other features of madvr are processed correctly?

And if i disable the DX11, into the video, when i press CTRL+J, madvr show 8bit winodowed mode instead 10bit

[Asmodian]

Ouch! Are you using native dxva2 decoding? Is it an Optimus system?

If you want 10-bit for some reason you have to use DX11 and exclusive mode.

madVR currently does not support sending HDR meta data and it doesn't​ matter what mode you use, HDR is always handled the same way.

[nevcairiel]

madVR currently does not support sending HDR meta data and it doesn't​ matter what mode you use, HDR is always handled the same way.

Actually, that was added in v0.91.9 (https://forum.doom9.org/showthread.php?p=1804847#post1804847). It can now send HDR metadata through two different methods, the NVIDIA API, which works on all OSes and with any rendering mode (but requires an NVIDIA GPU, obviously), and the Windows 10 API, which requires Windows 10 creators update and DX11 presentation mode.