madVR - high quality video renderer (GPU assisted)

[huhn]

as far as i know repeated or dropped frame forced by vsync are not shown in the OSD.

[Mark Regalo]

Originally Posted by huhn

as far as i know repeated or dropped frame forced by vsync are not shown in the OSD.

How can we solve problems related to vsync dropped/repeated frames if it's not shown in the OSD? Thanks.

[huhn]

SM or match your frame rate. custom resolutions is a possibility too.

[Xaurus]

Hi Guys,

I've changed my HTPC from this:
Silverstone GD08B/Intel 2120T/Noctua NH-U9BSE2/Asus P8H67-M Pro/Corsair XMS 2x2GB/Samsung 830 256 GB/7x2TB HDD/Asus GTX 660/Seasonic X-400 to the setup currently in my signature.

So my question is, is there somewhere in this massive thread there is a good explanation of NNEDI and frame doubling and stuff like that?

With my previous setup I used Jinc 3 tap but now I am not sure what is optimal with the considerably more powerful GTX 770.

Thanks.

[Anime Viewer]

Originally Posted by Xaurus

So my question is, is there somewhere in this massive thread there is a good explanation of NNEDI and frame doubling and stuff like that?

With my previous setup I used Jinc 3 tap but now I am not sure what is optimal with the considerably more powerful GTX 770.

Thanks.

Opinions are varied on the usefulness of NNEDI3 in its three enable-able areas as well as what settings to use.

I have NNEDI3 enabled for Luma image doubling, and have it set to 16 neurons since I can see a difference in anti-aliasing when I have it enabled compared to disabled, but I don't see a difference between 16 neurons and any of the higher settings with the content I watch. Some may recommend you use more neurons, but I say you decide that for yourself. Play a video, pause it, switch between the different settings (clicking apply after each) and see if you notice a change in picture quality for the better. If you do - use the new setting, but if not leave it at the lower setting and use those resources elsewhere.

madshi has said that he found during his testing that chroma doubling didn't have a significant effect, and was a huge hit on performance. As a result he doesn't recommend using it.

As far as chroma upscaling by NNEDI3 goes I don't notice any significant improvement using it over Jinc in the videos I watch, but again you can switch back and forth between the two while watching your videos and see if you see a difference. Another thing I noticed about NNEDI3 when used for chroma upscaling is that it can have a big hit to rendering speed even when it shouldn't be doing much (like if your playing a 1920x1080 video on a 1920x1080 screen. I posted the below on a different forum where I am attempting to help them determine presets for a codec pack that includes madVR using NNEDI3 options to be preset in their setup:

(1):
chroma upscaling -jinc 3 taps
luma double - always if upscaling is needed - 16 neurons
image upscaling - bilinear
downscaling - catmull-rom

(2):
chroma upscaling -jinc
luma double - always if upscaling is needed - 16 neurons
image upscaling - jinc
downscaling - catmull-rom

(3):
chroma upscaling -jinc
luma double - always if upscaling is needed - 32 neurons
image upscaling - bilinear
downscaling - catmull-rom

(4):
chroma upscaling -jinc
luma double - always if upscaling is needed - 32 neurons
image upscaling - jinc
downscaling - catmull-rom

(5):
chroma upscaling -NNEDI3 - 16 neurons
luma double - always if upscaling is needed - 16 neurons
image upscaling - bilinear
downscaling - catmull-rom

(6):
chroma upscaling -NNEDI3 - 16 neurons
luma double - always if upscaling is needed - 16 neurons
image upscaling - bilinear
downscaling - catmull-rom

(7):
chroma upscaling -NNEDI3 - 32 neurons
luma double - always if upscaling is needed - 16 neurons
image upscaling - bilinear
downscaling - catmull-rom

For people who mainly watch 480p videos the fastest order would probably be (fastest listed first, slowest last) option 1, 3, 5, 2, 7, 4, then 6.

480p video render speeds
option #1: 10ms
option #2: 20ms
option #3: 16ms
option #4: 23ms
option #5: 18ms
option #6: 25ms
option #7: 21ms

720p video render speeds
option #1: 22ms
option #2: 29ms
option #3: 27ms
option #4: 31ms
option #5: 31ms
option #6: 35ms
option #7: 33ms

1080p video render speeds
option #1: 7.5ms
option #2: 7.5ms
option #3: 9ms
option #4: 8.6ms
option #5: 24ms
option #6: 24ms
option #7: 34ms

As you can see with the 1080p video tests options 5,6,& 7 which use NNEDI3 for chroma upscaling take four times longer to render then Jinc 3 tap (options 1,2,3, and 4).

[Asmodian]

Originally Posted by Anime Viewer

Another thing I noticed about NNEDI3 when used for chroma upscaling is that it can have a big hit to rendering speed even when it shouldn't be doing much (like if your playing a 1920x1080 video on a 1920x1080 screen.

Actually chroma upscaling should be doing exactly the same work when playing a 1920x1080 video on a 1920x1080 screen as it would with any other resolution screen.

The chroma is 1/2 the resolution of luma in both directions (4:2:0). You need to resize it to match the luma before converting to RGB. This means that chroma scaling is always used and is independent of the output resolution.

This idea seems to be hard for some:
There are three image planes for a color image. On a bluray the black and white information (luma, Y) is stored at 1920x1080. Color information is stored as two 960x540 images (chroma, U & V or Cb & Cr). One needs to resize the two 960x540 images to match the 1920x1080 luma before converting to RGB.

This resize happens before anything else madVR does and it almost always needs to be done (it is only not needed when watching a 4:4:4 or RGB video).

[tjcinnamon]

Are there particular settings I need to set to get 4:4:4 RGB. My videos are always at 4:2:0.

Is that due to panel constraints?

[Asmodian]

Originally Posted by tjcinnamon

Are there particular settings I need to set to get 4:4:4 RGB. My videos are always at 4:2:0.

Is that due to panel constraints?

madVR always outputs 8-bit RGB, no setting needed. RGB cannot have subsampled color (Red, Green, Blue) so 4:4:4 or 4:2:0 do not apply.

Your display might do anything to the RGB or your GPU might convert to YCbCr if set to.

[tjcinnamon]

Originally Posted by Asmodian

madVR always outputs 8-bit RGB, no setting needed. RGB cannot have subsampled color (Red, Green, Blue) so 4:4:4 or 4:2:0 do not apply.

Your display might do anything to the RGB or your GPU might convert to YCbCr if set to.

Doesn't MadVR show the output of the video in the statistics and not the displays output?

Meaning that if it says 4:2:0 -> 4:2:0 that is what my receiver is getting and if I changed a setting to 4:2:0 to 4:4:4 or RGB then even if my display downconverted it wouldn't matter due to MadVR only record what's it output is?

[detmek]

MadVR always outputs RGB. 4:2:0 -> 4:2:0 is decoder output.

[johnniedoo]

Originally Posted by Anime Viewer

Opinions are varied on the usefulness of NNEDI3 in its three enable-able areas as well as what settings to use.

I have NNEDI3 enabled for Luma image doubling, and have it set to 16 neurons since I can see a difference in anti-aliasing when I have it enabled compared to disabled, but I don't see a difference between 16 neurons and any of the higher settings with the content I watch. Some may recommend you use more neurons, but I say you decide that for yourself. Play a video, pause it, switch between the different settings (clicking apply after each) and see if you notice a change in picture quality for the better. If you do - use the new setting, but if not leave it at the lower setting and use those resources elsewhere.

madshi has said that he found during his testing that chroma doubling didn't have a significant effect, and was a huge hit on performance. As a result he doesn't recommend using it.

As far as chroma upscaling by NNEDI3 goes I don't notice any significant improvement using it over Jinc in the videos I watch, but again you can switch back and forth between the two while watching your videos and see if you see a difference. Another thing I noticed about NNEDI3 when used for chroma upscaling is that it can have a big hit to rendering speed even when it shouldn't be doing much (like if your playing a 1920x1080 video on a 1920x1080 screen. I posted the below on a different forum where I am attempting to help them determine presets for a codec pack that includes madVR using NNEDI3 options to be preset in their setup:

(1):
chroma upscaling -jinc 3 taps
luma double - always if upscaling is needed - 16 neurons
image upscaling - bilinear
downscaling - catmull-rom

(2):
chroma upscaling -jinc
luma double - always if upscaling is needed - 16 neurons
image upscaling - jinc
downscaling - catmull-rom

(3):
chroma upscaling -jinc
luma double - always if upscaling is needed - 32 neurons
image upscaling - bilinear
downscaling - catmull-rom

(4):
chroma upscaling -jinc
luma double - always if upscaling is needed - 32 neurons
image upscaling - jinc
downscaling - catmull-rom

(5):
chroma upscaling -NNEDI3 - 16 neurons
luma double - always if upscaling is needed - 16 neurons
image upscaling - bilinear
downscaling - catmull-rom

(6):
chroma upscaling -NNEDI3 - 16 neurons
luma double - always if upscaling is needed - 16 neurons
image upscaling - bilinear
downscaling - catmull-rom

(7):
chroma upscaling -NNEDI3 - 32 neurons
luma double - always if upscaling is needed - 16 neurons
image upscaling - bilinear
downscaling - catmull-rom

For people who mainly watch 480p videos the fastest order would probably be (fastest listed first, slowest last) option 1, 3, 5, 2, 7, 4, then 6.

480p video render speeds
option #1: 10ms
option #2: 20ms
option #3: 16ms
option #4: 23ms
option #5: 18ms
option #6: 25ms
option #7: 21ms

720p video render speeds
option #1: 22ms
option #2: 29ms
option #3: 27ms
option #4: 31ms
option #5: 31ms
option #6: 35ms
option #7: 33ms

1080p video render speeds
option #1: 7.5ms
option #2: 7.5ms
option #3: 9ms
option #4: 8.6ms
option #5: 24ms
option #6: 24ms
option #7: 34ms

As you can see with the 1080p video tests options 5,6,& 7 which use NNEDI3 for chroma upscaling take four times longer to render then Jinc 3 tap (options 1,2,3, and 4).

I appreciate the details -particularly rendering times viz a viz each config.
Very helpful despite different set ups we have. It is good to see what others may experience with their trial and error techniques.
I do precisely the same thing with my videos though any 1920x1080 movies I watch I sometimes just use x64 MPC-BE svn and custom renderer. I find that MadVr offers fantastic PQ enhancement with my favorite types of old b&w videos, color,too, though I watch them with far less frequency just because there were fewer made. I am new to using MadVr and MPC-BE together so still limited in my scope. Never used this combo on any BD disc or iso, not even the mkv versions i make. not yet.
Thanks again for the information
John

[tjcinnamon]

Originally Posted by detmek

MadVR always outputs RGB. 4:2:0 -> 4:2:0 is decoder output.

Ahhhhh... So is there a way to increase the decoder output? I checked LAV and I have everything set.

So madvr outputs RGB and LAV is down converting it (for some reason)

[huhn]

no BD/DVD is RGB and Lavfilter doesn't down convert with a normal source.

[jmonier]

Originally Posted by tjcinnamon

Ahhhhh... So is there a way to increase the decoder output? I checked LAV and I have everything set.

So madvr outputs RGB and LAV is down converting it (for some reason)

The decoder (LAV) ALWAYS comes before the video renderer (madVR). That's basic to understanding how things work.

[tjcinnamon]

Originally Posted by jmonier

The decoder (LAV) ALWAYS comes before the video renderer (madVR). That's basic to understanding how things work.

How can I make these stats: http://i.imgur.com/bdjT3.jpg

look like these stats: http://i155.photobucket.com/albums/s...ScreenCut2.png

Or is that not possible?

[huhn]

this stats says that the source is 8 bit 4:2:0 is send as NV12

the screen from the 2nd screen is a wrong setup and not worth achieving.

[Asmodian]

Right, the second screen is doing a lower quality conversion to RGB in LAV instead of letting madVR do it.

[DarkSpace]

Originally Posted by Asmodian

Right, the second screen is doing a lower quality conversion to RGB in LAV instead of letting madVR do it.

Not even in LAV, but in ffdshow... and I find it interesting that madVR specifies that the fullrange information comes not from upstream, but specifically from ffdshow, too.

[detmek]

Originally Posted by tjcinnamon

Ahhhhh... So is there a way to increase the decoder output? I checked LAV and I have everything set.

So madvr outputs RGB and LAV is down converting it (for some reason)

Yes, but don't do that. madVR does better job converting YV12 or other formats to RGB.

As already explaind, playback chain goes: splitter > decoder > render.

So, let the LAV Splitter and LAV Decoder decode your file without any conversion (YV12 to NV12 is not conversion, those are same format, different order) and madVR do any scaling and colorspace conversion.

Don't try to overcomplicate things. Just because something looks messy and complicated (most online tutorials) does not make it better.

[Anime Viewer]

Originally Posted by detmek

Yes, but don't do that. madVR does better job converting YV12 or other formats to RGB.

As already explaind, playback chain goes: splitter > decoder > render.

So, let the LAV Splitter and LAV Decoder decode your file without any conversion (YV12 to NV12 is not conversion, those are same format, different order) and madVR do any scaling and colorspace conversion.

So, what are you advocating? That he/she go into LAV video decoder and uncheck all the RGB options, so that LAV doesn't convert anything to RGB?

Originally Posted by tjcinnamon

How can I make these stats: http://i.imgur.com/bdjT3.jpg

look like these stats: http://i155.photobucket.com/albums/s...ScreenCut2.png

Or is that not possible?

I didn't view your second file, but theoretically if you want your 4:2:0 videos to be converted to 4:4:4 videos you could go into LAV video decoder, and uncheck all the output formats except for the 4:4:4 ones. It will not necessary lead to a better picture (it might under some circumstances), and it will use LAV to do conversion (which some people here don't like). Regardless you could test with it, and see if you notice any change in quality (good or bad). While you're in LAV video decoder proprieties you could also change RGB output levels to Untouched which would eliminate LAV as a conversion factor for that if nothing else.

If you're concerned that your colors look washed out in your videos you can experiment with adding the 16-235 to 0-255 [SD][HD] to active shaders in MPC under options -> Playback -> Shaders. Some people don't like it feeling that they loose some of their descreet greys while others feel it adds vibrancy.