madVR - high quality video renderer (GPU assisted)

[huhn]

smooth motion is as good as it is possible to do. it is not meant for frame interpolation it is meant to remove judder from frame rate mismatch.

[mclingo]

I am told smooth motion looks better at higher TV/monitor refresh rates, sadly my TV ancient and only supports 60hz, havent wanted to upgrade it 3D Passive OLED is still a better than cinema experience, after 10 years i'm sure i've lost some brightness though.

[huhn]

yes the blended frames are only shown for one vsync with higher refreshrate the vsync is shorter resulting in less chance to see blending artifacts.

[Amuat]

I wonder if there's something wrong with my setup as I've yet to see any visible difference whatsoever from 'reduce ringing artifacts', testing it multiple times during scenes where it was sorely needed. Are there some settings that can block it? All the other options on the artifact removal tab work wonders (never really realized the power of reducing banding artifacts until testing it with a space documentary), and were the deciding factor for changing back from MPCVR despite all the incompatibility issues. So that is what I would love to see taking its rightful place among its esteemed peers.

[huhn]

should always work and works just fine:
https://slow.pics/c/cuhsRlwC

[VergiL]

The 10bit full screen problem seems to have disappeared for me... to be confirmed.
Edit: no, the problem is still here.

[huhn]

clearly broken here.

[Amuat]

Since even MPCVR doesn't make use of it (AFAIK), is it very difficult to take advantage of DX12, or is there just not much to be gained from it? Are there (other?) aspects from newer hardware that hold clear potential but are yet to be utilized?

[huhn]

there quite an old port of it for dx12 but yeah what do you expect from dx12 for video?

[Sunspark]

It's theoretically possible it might make a compatibility difference with the video drivers or the system. Win 10 moved the compositor and dwm to dx12 with 2004.

[huhn]

not that it matter but win 11 uses legacy flip pretty sure that is not DX12.

[Amuat]

there quite an old port of it for dx12 but yeah what do you expect from dx12 for video?

Well I don't expect ray-tracing, but I thought it was reasonable enough to wonder whether it could bring some kind of improvements as DX11 has enabled advances in both renderers if I'm not mistaken. But I guess then that 11 had broader applicability whereas 12 is more exclusively beneficial to just gaming?

[huhn]

i'm not aware of a single video rendering advantage of DX12 and yes dx11 has them over dx9 like it can do 10 bit presentation natively but madVR is DX9 it only does very limited stuff in DX11.

are newer hlsl level useful yeah maybe... we may learn about that in 5 years or so.

in mpv dx11 performance is usually better then vulkan but vulkan is very comparable to dx12 and focus on vulkan makes still sense here because it is crossplattform.

[ashlar42]

I don't think anybody actually knows any of the plans for MADVR From the software point of view, we are still just being used as guinea pigs and testers for Madvr Envy which is perfectly fine because it's still free. Obviously we'd all like to see some more development I personally would like to see Some more work done on smooth Motion to make it usable.
Thank you for the answer.

[amayra]

is there any new version of madvr

since developer stop posting here and can't keep up with new version when it released

[mclingo]

is there any new version of madvr

since developer stop posting here and can't keep up with new version when it released


you do know google exists?

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

[VergiL]

After about an hour of watching, glitches appear every second. This is due to the composition rate, which suddenly switches to 23.000Hz during playback instead of 23.976Hz. If I pause the video, the composition rate resets to 23.976Hz, and as soon as I start playback, it goes back to 23.000Hz. You then have to either stop the movie or move forward or backward using the progress bar. Any idea what could be triggering this?

[dead_seth]

I want the BEST picture quality from madvr.
Do you think 9070XT or 3090 will be better ?
Also im using potplayer do u think mpc-hc/be, mpv or anyone else is better?

Im watching mostly 4k movies on 1440p monitor and sometimes on 4k tv.

[Sunspark]

Best doesn't exist because different algorithms produce different results.

[dead_seth]

What about the gpus??
3090 or 9070xt will be more powerfull for madvr?

Now with vega64 i have 35-45ms render time with drops ang glitch but they are not visible.. i know its too high.
Thats why i ask abouyt 3090 or 9070xt

[Amuat]

As I mentioned in an earlier post, I use a distinct, as neutral as possible, profile for my own footage, and was wondering if SoftCubic100 would be more neutral than Jinc (with anti-ringing filter) for chroma? It adds barely any sharpening, only slight aliasing (which for some reason lessens with anti-ringing filter off, which isn't at all needed anyway if the bars are to be trusted). The thing is that it apparently hides source artifacts in a massive way. What could that entail exactly? If I ever actually want to see any artifacts that are part of the source material it's precisely when checking my own footage.

[Sunspark]

I use Mitchell-Netravali AR in the cubic drop-down. Try that.
Bicubic 60 AR is also good too.

All the items in that cubic list are the same formula, just with different values for B and C.

I use M-N because it doesn't require system resources and provides a naturalistic look. Other algorithms can make some colours too saturated for my liking.

[huhn]

As I mentioned in an earlier post, I use a distinct, as neutral as possible, profile for my own footage, and was wondering if SoftCubic100 would be more neutral than Jinc (with anti-ringing filter) for chroma? It adds barely any sharpening, only slight aliasing (which for some reason lessens with anti-ringing filter off, which isn't at all needed anyway if the bars are to be trusted). The thing is that it apparently hides source artifacts in a massive way. What could that entail exactly? If I ever actually want to see any artifacts that are part of the source material it's precisely when checking my own footage.

what bars?
and no jincs is so soft it massively washes out colors i just did a lot of RGB to 420 and back to 444 tests testing a new unnamed scaler.

even super xbr 150 is washing out color and is unnatural.
biliteral is the "worst" scaler available but it can rarely archive true native results but it more often then not just messes the image up.
super xbr 150 AR and NGU AA very high (sadly matters...) are by far the best scaler.

softcubic is for bit rate starved mpeg2 encodes it starts to hide the blocking for chroma i just doubt that.
I use Mitchell-Netravali AR in the cubic drop-down. Try that.
Bicubic 60 AR is also good too.

All the items in that cubic list are the same formula, just with different values for B and C.

I use M-N because it doesn't require system resources and provides a naturalistic look. Other algorithms can make some colours too saturated for my liking.
you may like it you may think it is natural but bicubic results are just objectively wrong.

test true RGB images.

[nautilus7]

Hi, I haven't been following this thread, so it might have been discussed before...

Since I move to windows 11 a year ago, madvr is not able to display any HEVC content properly. The picture looks like in the image below. On the contrary mpc-hc with its default renderer plays fine.

https://i.ibb.co/HDMtvz7k/madvr.png (https://ibb.co/HDMtvz7k)

[Aleksoid1978]

Just don't use madVR )

[mclingo]

Hi, I haven't been following this thread, so it might have been discussed before...

Since I move to windows 11 a year ago, madvr is not able to display any HEVC content properly. The picture looks like in the image below. On the contrary mpc-hc with its default renderer plays fine.

https://i.ibb.co/HDMtvz7k/madvr.png (https://ibb.co/HDMtvz7k)

I have no issues with my windows 11 and amd card like this what Have you done to try and solve this yet, Have you done a full driver wipe with which had the type of card you are using AMD or NVIDIA have you switched from d3d11 to dxva etc

[Amuat]

what bars?

The ones on the top-right of the window; sharpness, hide source artifacts, aliasing...

How much exactly does it matter that AA is very high instead of high or even medium? I'm shifting the discussion to other profiles now, but regarding image upscaling with NGU, are there other differences than just sharpness between the different variants?



I use Mitchell-Netravali AR in the cubic drop-down. Try that.
Bicubic 60 AR is also good too.

Both of those seem to have a lot of aliasing going on. Or are those bars just mostly irrelevant?

[Sunspark]

I wouldn't say irrelevant, but I don't have information on the veracity of them.

They are useful in the sense that it helps identifying luma scalers that don't have a sharpness that is very different from the chroma scaler's sharpness.

[nautilus7]

I have no issues with my windows 11 and amd card like this what Have you done to try and solve this yet, Have you done a full driver wipe with which had the type of card you are using AMD or NVIDIA have you switched from d3d11 to dxva etc

Hi, I had tried different settings. And sometimes AVC video did not play as well.

Now, I tried disabling "use direct3d11 for presentation" in madvr settings and voila, it works! Thanks for the hint!

Btw, I am using 12th gen intel igpu for graphics.

[huhn]

The ones on the top-right of the window; sharpness, hide source artifacts, aliasing...

How much exactly does it matter that AA is very high instead of high or even medium? I'm shifting the discussion to other profiles now, but regarding image upscaling with NGU, are there other differences than just sharpness between the different variants?





Both of those seem to have a lot of aliasing going on. Or are those bars just mostly irrelevant?

ignore these.

I wouldn't say irrelevant, but I don't have information on the veracity of them.

They are useful in the sense that it helps identifying luma scalers that don't have a sharpness that is very different from the chroma scaler's sharpness.

chroma is double scaled so this is a very flaws thinking about it.

doing 540 to uhd in 1 step or 2 steps using the same dumb scaler will result in massively different results the one where the scaler is used twice will look plain terrible in comparison

[huhn]

Just don't use madVR )

you know an alternative with proper SM 3D LUT and a chroma scaler that does not insult me?

i really need one (

[Siso]

you know an alternative with proper SM 3D LUT and a chroma scaler that does not insult me?

i really need one (

Well said, huhn.

[huhn]

is there any new version of madvr

since developer stop posting here and can't keep up with new version when it released

you do know google exists?

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

it's 208 that page only has 207 at the time of posting.

so here you go http://madshi.net/madVRhdrMeasure208.zip
let me see if i can force a 209 today.

[Amuat]

ignore these.

You mean the whole thing you quoted or just the bars? Meaning NGU soft and sharp for example differ in no way other than sharpness? And the AA chroma: very high is significantly better than high then..?

[huhn]

You mean the whole thing you quoted or just the bars? Meaning NGU soft and sharp for example differ in no way other than sharpness? And the AA chroma: very high is significantly better than high then..?

ignore the bars. it does not really apply to chroma but i'm not saying they are wrong lanczos is sharper then bicubic 50.

for NGU very high is always better then high.
it is not sharper the neural network is bigger and such making "better" decision.

NGU soft standard and sharp are as far as i can see the same algorithm with a different sharpness. my best guess is that soft and standard are just there for low quality sources but they are still not good there. the king is NGU AA anyway.
even sharp is softer then original downscaled image even when you use a sharp downscaler and these scaler are trained on bicubic.

and yes NGU AA very high can look massively different and better to NGU AA high for chroma. that doesn't mean you should use it it is still chroma what you are scaling and using a ton of resources for it is not that wise.

NGU sharp standard and soft have no place in chroma i can't remember a single test where it wins the kings are super xbr and NGU AA.

[Siso]

Now that we cleared the chroma thing with super xbr 150 and NGU AA very high, how about the luma situation?

[Sunspark]

Still on the side that says there is no one perfect algorithm. If there was, it would be the only one in use.

The content you watch makes a difference too.. e.g. anime vs regular scenery.

[huhn]

NGU AA works with pretty much everything that is no pixel art same for super XBR

NGU sharp can't handle compression artifacts that's one of the reason you can run it with artifact removal because it really can't handle them and it shines for bicubic downscaled sources because that's what it is trained for undoing a bicubic downscale as good as possilbe.

just use NGU AA is very very save to use.

[Siso]

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

[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.

[huhn]

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]

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]

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]

Does it affect the Aspect Ratio of the files?
no it does not affect that at all.

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.

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.
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]

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 (https://forum.doom9.org/showthread.php?p=1709584) 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) (https://prnt.sc/xOh2_6o6FLwP)

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