madVR - high quality video renderer (GPU assisted)

[huhn]

Originally Posted by Siso

NGU AA for luma? It moves the pixels by 1, unlike any other scaler.

it's 0.5 just like nnedi3 pretty much just an update for that scaler.
nnedi is a deinterlacer not a scaler so that why it moves the image top field and bot field and such. ngu aa does the same so you can easily assume it also works like a bob deint algorithm.

and because it moves the image by half a sub pixel it can't be compared with metrics.

[Siso]

Originally Posted by huhn

it's 0.5 just like nnedi3 pretty much just an update for that scaler.
nnedi is a deinterlacer not a scaler so that why it moves the image top field and bot field and such. ngu aa does the same so you can easily assume it also works like a bob deint algorithm.

and because it moves the image by half a sub pixel it can't be compared with metrics.

Does it affect the Aspect Ratio of the files?

[Amuat]

So AA is your recommendation for luma too? I mean, like said earlier, content is a factor too, but if you had to choose one for general use? Is the difference here as significant between very high and high as with chroma?

[Asmodian]

I prefer NGU AA over the others for general content.

The performance hit for using NGU AA very high for chroma scaling is lower because it processes half the number pixels (it is used during the conversion from 4:2:0 to 4:4:4).

However, luma is always more important than chroma. The difference between very high and high is also more important for luma. There is more detail in the luma to work with and humans are more sensitive to luma details. You are much more likely to notice a difference between NGU high and very high with luma scaling than chroma scaling.

[Sunspark]

Originally Posted by GoldenSW

1. I have some older 576i interlaced videos and want madVR to handle the deinterlacing for maximum quality, of course only if it's better than LAV at deinterlacing.

Just leave LAV settings at no software deinterlacing, and adaptive hw unchecked.

In MadVR, check disable automatic source type detection and select "force film mode". This is for stuff that was filmed at 24 fps and aired at 30 fps on 60 Hz displays which is almost everything before they switched over to digital filming. After, it depends.

Hardware Decoder: Given my hardware and the use of madVR, what is the best option under "Hardware Decoder to use" in LAV Video? Is D3D11 generally preferred over DXVA2?

You can use D3D11 Automatic decoding and have D3D11 presentation mode checked (which is an interop to the underlying D3D9 code anyway). My Windows setup is old, so I just use copy-back dxva2.

[JNW]

Strange. I per prefer no pixel offset.

[huhn]

Originally Posted by Siso

Does it affect the Aspect Ratio of the files?

no it does not affect that at all.

Originally Posted by GoldenSW

Hi everyone,
I'm trying to fine-tune my MPC-HC setup for the best possible video quality and performance using madVR and I have a couple of questions. Any help would be greatly appreciated!
My setup: MPC-HC 2.4.1, madVR build 207, fully updated Windows 11, RTX 2080, i7-9700k

1. I have some older 576i interlaced videos and want madVR to handle the deinterlacing for maximum quality, of course only if it's better than LAV at deinterlacing.
In LAV Video Decoder's "Deinterlacing Mode" settings, what's the recommended approach? Should I select "Disabled (Progressive)" to ensure LAV doesn't touch it? Or is setting it to "Auto" generally sufficient to let madVR take over when needed? If "Auto" is selected, and the "Software Deinterlacing" sub-option is "No Software Deinterlacing", does this guarantee LAV won't interfere, or could Auto still potentially perform deinterlacing itself sometimes?

2. Optimal hardware acceleration settings:
Hardware Decoder: Given my hardware and the use of madVR, what is the best option under "Hardware Decoder to use" in LAV Video? Is D3D11 generally preferred over DXVA2?
Hardware Device: I understand madVR often needs a "copy-back" mode to allow its processing (including deinterlacing). Are there any scenarios when using madVR where a "native" mode is recommended or safe since these is a noticeable performance difference, or should I always stick with a copy-back mode to ensure madVR works correctly? It may be worth noting I don't use anything in the "processing" category in madvr except very rarely deinterlacing.

the best hardware decoder for interlaced content is by far >none<
hardware decoder can turn soft telecine to hard telecine and you really don't want that.

the lavfitler default settings for deint are perfection.

keep your hands of d3d11 native decode it is digital satan and needs to be punished by a hammerhead to the face in the name of ludd thanks to madVR not using it as it should be used.

Originally Posted by Amuat

So AA is your recommendation for luma too? I mean, like said earlier, content is a factor too, but if you had to choose one for general use? Is the difference here as significant between very high and high as with chroma?

yes NGU AA is tha best and da greatest.

Originally Posted by Asmodian

I prefer NGU AA over the others for general content.

The performance hit for using NGU AA very high for chroma scaling is lower because it processes half the number pixels (it is used during the conversion from 4:2:0 to 4:4:4).

However, luma is always more important than chroma. The difference between very high and high is also more important for luma. There is more detail in the luma to work with and humans are more sensitive to luma details. You are much more likely to notice a difference between NGU high and very high with luma scaling than chroma scaling.

for some unknown reason NGU AA very high can grow wings when chroma scaling showing ridiculous difference between very high and high.

even with that case your core statement is obviously correct luma is so much more important that this fact does not matter much if at all.
but very high is not worth the performance cost for luma.
and general i think NGU AA is not worth it for chroma it's to expensive but that is my problem.

i saw things i can't even bring to words.

[GoldenSW]

Originally Posted by huhn

no it does not affect that at all.


the best hardware decoder for interlaced content is by far >none<
hardware decoder can turn soft telecine to hard telecine and you really don't want that.

the lavfitler default settings for deint are perfection.

keep your hands of d3d11 native decode it is digital satan and needs to be punished by a hammerhead to the face in the name of ludd thanks to madVR not using it as it should be used.

But does madvr use hardware deinterlacing?
Also do you have any concrete examples of why using native might be bad for my context?
I have found this information in this post:

-DXVA2 Native DX9 hardware decoding returns bad quality (blurred chroma) on Nvidia GPUs. The DX11 hardware decoding API (D3D11VA) that is used when LAV Video is set to D3D11 Automatic (so native decoding is used), and madVR is set to use Direct3D 11, avoids these problems. So do all copy-back hardware or software decoding methods.
-All Native hardware decoding methods result in some functional ramifications. It is impossible to use force film (IVTC) and the detection of black bars does not work. madVR can still shift the entire video (including any hard coded black bars) to the top or the bottom of the screen, as well as crop pixels off all sides, but it cannot detect black bars. Again, copy-back hardware or software decoding avoids these problems.

[huhn]

that's accurate and should be enough.
with D3D11 native madVR will tell you it can't deint.

madVR can do ivtc(only with copyback or software) or use a hardware deint provided by the GPU driver it has nothing else available. hardware deint is often broken. broken deint is now a feature on AMD for how long it is been broken.

film mode=IVTC
video mode=deint

edit:

-DXVA2 Native DX9 hardware decoding returns bad quality (blurred chroma) on Nvidia GPUs. The DX11 hardware decoding API (D3D11VA) that is used when LAV Video is set to D3D11 Automatic (so native decoding is used), and madVR is set to use Direct3D 11, avoids these problems. So do all copy-back hardware or software decoding methods.

that bold part is wrong even with d3d9 presentation the issue is fix.

the issue is dxva2 native uses RGB for the interop so it has to chroma scale and rgb convert which will do quite some damage and is well not bit perfect.
with d3d11 it can transfer the image natively as is so it is so bit perfect and the first thing madVR does is an d3d11 d3d9 interop which is also lossless (first line in the OSD).

[JohnRo79]

hi guys

it seems that when i use MadVR with Mpc-BE , when i go full screen, the lower dark bar and the top one
are both covering the Subtitles (bottom) and even the TV's native Display info (at the top)

using v 00.92.17 (forgot on how to check what build)

[huhn]

neither the PC nor madVR can black the TVs native info bar only the TV can.

but why would the time information survive...

[Asmodian]

Originally Posted by huhn

that bold part is wrong even with d3d9 presentation the issue is fix.

Thanks! Fixed.

[JohnRo79]

Originally Posted by JohnRo79

hi guys

it seems that when i use MadVR with Mpc-BE , when i go full screen, the lower dark bar and the top one
are both covering the Subtitles (bottom) and even the TV's native Display info (at the top)

using v 00.92.17 (forgot on how to check what build)

i went all nuts on it: ddu, new drivers, new madvr, new mpc-be, fresh settings
all is fine now

[purplebaby]

I'm unsure if HDR is working correctly. I've just upgraded from Win 10 to Win 11.

I'm using MPC-HC x64 latest version and madVR.

When I launched hdr movies with this setup in Win 10, my TV (LG CX, secondary display, connected via HDMI, I don't 'extend', I use only the monitor or only the TV at a time) would switch over to HDR mode, with colors matted in windowed mode, and with correct HDR display in fullscreen mode.

But now in Win 11, it appears to have somewhat correct colors in windowed mode as far as I can tell, perhaps without the fullest contrast hdr can give, but it does NOT switch the TV into hdr mode. It remains in standard mode regardless of fullscreen or not.

Rtx 5090 (572.83 drivers). Any advice would be appreciated.

[purplebaby]

I figured ut out. I had to remove displays from madvr, unplug the monitor cable and let madvr discover the tv afresh as the only connected display, making it primary.

But now I can't get madvr to NOT enable hdr on my IPS desk monitor... I've set it to tone mapping, but it still switches to hdr mode when opening a hdr file.

[huhn]

install the newest beta could be the win 11 device bug.

[GoldenSW]

@huhn, @Sunspark thank you! appreciate the input! One thing is for sure, there doesn't seem to be much advantage in using d3d11 native instead of copy-back with powerful modern hardware. Performance difference is maybe 1-1.5ms with my setup.

[purplebaby]

Originally Posted by huhn

install the newest beta could be the win 11 device bug.

Newest beta of what? I don't see any betas for madvr

[JohnRo79]

Originally Posted by purplebaby

Newest beta of what? I don't see any betas for madvr

https://www.videohelp.com/software/madVR

[purplebaby]

Originally Posted by JohnRo79

https://www.videohelp.com/software/madVR

Thanks.

That worked.

I still get the issue, sometimes, of HDR mode being entered into on the TV when opening a hdr file, but the colors are extremely matted and dimmed video, such as you typically see outside the video window. I have to reboot to get it back to working normally.