madVR - high quality video renderer (GPU assisted)

[ryrynz]

The ideal way to test chroma is DON'T BOTHER.



Waste of time, waste of Gpu power.



Impossible to notice from real viewing distance on real content, you can only see it on very specific test patterns.


That's a complete load of shite. There have been scenes I can clearly notice the difference in real world content.

[tp4tissue]

That's a complete load of shite. There have been scenes I can clearly notice the difference in real world content.

I'm not saying it doesn't exist, I'm saying such scenes are rare, and it's equally rare that you'd notice if you're not looking for it. :scared:

[.m4jX]

I'm not saying it doesn't exist, I'm saying such scenes are rare, and it's equally rare that you'd notice if you're not looking for it. :scared:

Yesterday you wrote that it's "impossible" to notice on real content. You get called out, now it's "I'm not saying it doesn't exist". I don't think you know what impossible means.

If you wanna keep ragging on huhn misunderstanding things you write, you might wanna take some English lessons. Because I would be absolutely flabbergasted if you are a native speaker, because half the time you seem to be contradicting your own bullshit and then gaslight when you get called out, pretending you were misunderstood.

[flossy_cake]

I have a hard time noticing the difference in chroma algorithms, except if the content has some baked-in chroma defect like the Family Guy DVD we were discussing earlier, or on computer generated test patterns which don't have an MTF like cameras, or on special patterns that generally aren't found in camera content.

On the other hand I made the drastic decision to sell my 4k Panasonic OLED in favour of my crappy old 768p plasma because of how bad the chroma upscaling was on the Panasonic (visible chroma aliasing on 55" screen @ 3.5m viewing distance). I suspect its chroma scaling was bugged in the firmware. Also I hated its upscaling in general — even with sharpness set to max it still looked like it was set to 0 for most pixels due to the nature of its sharpening algorithm where it only sharpened very high contrast edges. For example the contrast difference in the whites of an actor's eyes and their pupil/iris often wasn't above that threshold and upscaled facial features appeared always blurry to me.

[chros]

Interesting, which model was it?

[flossy_cake]

55fz950u

[chros]

So that was the last year without Dovi in 2018. Interesting what you said, everybody praised that one as well, just as usual :)

[ryrynz]

55fz950uDamn. Can't say I'm hating what I see on mine but I'm using MadVR the majority of the time anyway.

[flossy_cake]

So that was the last year without Dovi in 2018. Interesting what you said, everybody praised that one as well, just as usual :)

At the time I suspected it was because the reviewers were typically sending it 4:4:4 and bypassing the Panasonic's chroma upscaling. At the time I was watching mostly broadcast TV at 4:2:0 interlaced. Nowdays all my equipment is outputting 4:4:4 progressive.

[flossy_cake]

@ryrynz which HDMI ports do you use -- side or rear?


The rear facing HDMI slots are not full-fat full bandwidth but can accept 4K 4:2:0 at 50/60hz. The side facing HDMIs are full bandwidth 4:4:4 compatible

[tp4tissue]

I have a hard time noticing the difference in chroma algorithms, except if the content has some baked-in chroma defect

exactly right flossy,

I tested myself many times on the chroma when nnedi was first "all the rage." on 1080p and above, I got it wrong 50% of the time with normal content.

The visual difference is incredibly subtle but the processing cost is extreme.

There may well be a correct/ better chroma, but the fact that we're still using 4:2:0 is proof that we're just not good at spoting it. :D

I'm all for 4:4:4, but eh... if there are other priorities, maybe later.

[70MM]

Is there any other users here who use Jriver MC29 with madvr?
If so have you ever had an issue bringing up the madvr stats using cntrl J?
I have issues bringing up the stats from time to time but never did on earlier JR players MC25 and MC21.
There is something about madvr loosing focus and a left mouse click returns the stats using cntrl J, this shouldnt be needed....

[ryrynz]

@ryrynz which HDMI ports do you use -- side or rear?

Side. Am aware of the limitation so makes sense.
I recently connected a new receiver to HDMI 2 and was wondering why I couldn't set RGB for the color format *oh yeah need to use HDMI 1* :p

[jkauff]

Is there any other users here who use Jriver MC29 with madvr?
If so have you ever had an issue bringing up the madvr stats using cntrl J?
I have issues bringing up the stats from time to time but never did on earlier JR players MC25 and MC21.
There is something about madvr loosing focus and a left mouse click returns the stats using cntrl J, this shouldnt be needed....
I think this began when JRiver integrated nevcariel's JRVR renderer into MC. It also uses Ctrl-J to bring up its OSD, and that key combination used to do something else in MC so they modified the code slightly. It's a bit of a pain, but if you give the display window focus first Ctrl-J seems to work as expected.

[Balling]

Supposedly 10bit encodes better than 8bit, even though the source is only 8bit. I don't know if it's necessarily true, but some of my 2mbps 10bit x265 1080p files look surprisingly decent for such little data. Grain is suppressed though, and for some reference movies I'm not satisfied.

Bigger problem seems to be storage — if I want a large collection at high bitrate / remux, I cannot find a HDD large enough and quiet enough to hold it all.

Yes, you need 10 bit YCbCr to just preserve 8 bit RGB source. Because YCbCr is bigger than RGB, so more bits are needed to preserve smaller part of the space.

[Balling]

should i mention UHD BD are 10 bit and p010 is absolutely fine?
or maybe i hope he just get bored and leaves?

I know that, so nonHDR UHD BD are also 10 bit and BT.2020 primaries. And of course it all works now, P010 and maybe even better. Still it was not so before 375.xx driver of Nvidia.

In Intel even better does not work...

[dupcengier27]

Why does PotPlayer on-screen display show a very large number of frames per second in the material, while the smooth motion option is turned off ...
https://i.ibb.co/T2Prr9G/wysoka-liczba-klatek-OSD-Pot-Player.png
https://i.ibb.co/bQ81Cjj/wysoka-liczba-klatek-OSD-Pot-Player-Madshi-VR.png
What are the recommended settings for this ATI Radeon HD4350 graphics card? because, for example, in the 1080p chapter, frames per second are not lost too much, but there is a significant number of presented glitches ...... does it make sense to use Madshi Video Render on this graphics card and this processor 2-core AMD Athlon X2 5200+ ? I noticed that when I changed image upscaling and downscaling to DXVA2 method, it is a dark image ......now I changed to Bilinear method, but here there is a certain number of lost frames and presented glitches......

[huhn]

there is some odd stuff happening here.

first of all the OSD says it is decoded to NV12 which is normal but madVR get's YV12. is NV12 disabled in lavfilter?

you are using D3D11 for presentation which is slightly slower and with present frames and delay until queue are full it may come up with bad numbers i would ignore that.

the GPU is most like not able to use madVR properly but you can try it with d3d9 presentation and decode or odd combination like lavfilter limited RGB output to offload some processing to the CPU.

presentation glitches are windows issues if you have win 7 you could try fullscreen exclusive with D3D9 where presentation glitches should be "impossible" as long as you are in full screen.

else try this:
https://github.com/Aleksoid1978/VideoRenderer/releases/tag/0.5.6
the thread for it is here:
https://forum.doom9.org/showthread.php?t=165890

[dupcengier27]

@ thanks for your reply huhn
this is the LAV Filters settings
https://i.ibb.co/5c5bMz9/LAV-Filters-ustawienia.png
..... what should I change to improve Madshi Video Render? Where should I change the presentation to d3d9 or use lavfilter limited RGB output to offload some processing to the CPU? I unchecked NV12 decoding in LAV Filters, but still the PotPlayer OSD shows that the frame rate is higher than the source material has ..... what could be the cause of this condition ....
also in the moment of pause, the PotPlayer OSD shows a very large frame, also in fullscreen exclusive mode ...... I get the impression that it is a desktop frame, not the movie or series itself...but in fullscreen exclusive mode screen came out dark ....
both during movie / series playback and pausing OSD show high frames.
https://i.ibb.co/Gn10LtJ/Hobbit-OSD-wysoki-klatka-w-momencie-pauzy.png
https://i.ibb.co/wMt7qzK/Blade-Runner-OSD-Madshi-ustawienia.png
I marked the YV12 and NV12 Filters in LAV and also during the pause there is a very large number of frames with YUY2 at the input ....
https://i.ibb.co/sjhchms/Hobbit-OSD-wysoki-klatka-w-momencie-pauzy-bez-NV12-YV12.png
current settings in LAV Filters.
https://i.ibb.co/41j2kk1/LAV-Filters-ustawienia-aktualne.png
but I do not feel the effect that a given movie or series would actually have such a large number of frames, this is a distortion in this OSD PotPlayer (incorrect reading of frames from the movie / series only from the Windows desktop as i suspect), especially that in the case of the standard Enhanced Video Render or MPC Video Render the correct number of frames is shown, only when Madshi Video Render is running, the OSD shows such a high number of frames ... ... ...
do you recommend DXVA2 or Bilinear method for upscaling and downscaling in Madshi ?
what a miracle the OSD Madshi shows that I am using D3D11 for the presentation, since ATI Radeon HD4350 like Windows 7 do not support this mode ..... where should I change to D3D9?
Why is a dark image displayed in 1080p content with such settings as in the screenshot attached? of course the sound is heard .....
https://i.ibb.co/T0TjLzt/Hobbit-OSD-Madshi-ustawienia.png
https://i.ibb.co/myK9VqS/Hobbit-OSD-Madshi-ustawienia-v.png
I apologize in advance for the correctness and readability of my entries, but I use translator as my written and spoken English is at a very low level ....

[tp4tissue]

the composition rate is not the file frame rate. it's 60 because that's the monitor refresh rate you're using. if you switch the monitor output to 23.967 or 24hz then composition rate will drop to that.

you can use bilinear/cubic for chroma, and lanzos or jinc for image upscaling.

These are nice and not oversharp, they're also low resource so that your card fan doesn't spin up too loud during the movie.

[flossy_cake]

Where should I change the presentation to d3d9

MadVR settings > rendering > general settings > untick "use Direct3D 11".

And try "enable automatic fullscreen exclusive mode" as that should do the best job to prevent dropped frames.

Also you might consider increasing your monitor refresh rate slightly to around 59.94hz to improve the "1 frame drop every 46.22 seconds" on your Ctrl+J screenshot. Can be done with either:


AMD GPU control panel custom resolution
MadVR's built in tool (devices > your display > display modes > custom modes)
Third party tool CRU (I'm using this)

Top line of Ctrl+J is the meaured refresh rate, or check it at vsynctester.com.


DXVA2 or Bilinear method for upscaling and downscaling in Madshi ?

Bilinear is a bit crap -- noticeably softer than Cubic & Lanczos which are pretty standard & I'd be surprised if your GPU couldn't manage it. Then again your GPU is a $40 card from 2008 so maybe it can't, I don't know.

The specs say it can decode MPEG-2, h.264 and VC-1, so you can tick those boxes in LAV Video Decoder and try it out to save some CPU, but because it's so old it might not decode correctly and you might be safer disabling it in LAV and letting CPU do the decoding.


Why is a dark image displayed in 1080p content with such settings as in the screenshot attached?

The screenshot doesn't look dark to me, but I noticed in LAV Video decoder you have RGB output levels set to "untouched" and I've got mine set to "PC (0-255)" so maybe try that. It's possible the screenshot isn't capturing what you're seeing on screen due to your configuration of MadVR settings > rendering > screenshots.

In any case use a test pattern (eg (https://drive.google.com/file/d/0B9FgjTtNzSy0dTZwZ3p1eGVnTG8/edit?usp=sharing)) to check your near-black and near-white tones.

[huhn]

@ thanks for your reply huhn
this is the LAV Filters settings
https://i.ibb.co/5c5bMz9/LAV-Filters-ustawienia.png
..... what should I change to improve Madshi Video Render? Where should I change the presentation to d3d9
ignore the potplayer OSD.

the output mode can be set under rendering -> general settings -> use direct3d 11.



I unchecked NV12 decoding in LAV Filters, but still the PotPlayer OSD shows that the frame rate is higher than the source material has ..... what could be the cause of this condition ....

do you recommend DXVA2 or Bilinear method for upscaling and downscaling in Madshi ?
DXVA scaling is buggy and the quality is up to the GPU driver it's known how good or bad it it expect bicubic.
what a miracle the OSD Madshi shows that I am using D3D11 for the presentation, since ATI Radeon HD4350 like Windows 7 do not support this mode
it's not a miracle at all madVR uses very basic d3d11 features the card supports it.

the HD4350 has less then 0.010 tera flops everything is very unlikely to work.
madVR in general is most likely not a good choice here.

use lavfilter limited RGB output to offload some processing to the CPU?
you need copyback decoding or software decoding(software is nearly for sure better because old AMD cards hate copy back operations) to do that and you have to disable everything except RGB24 if that doesn't work you can try RGB32. to make sure it is limited range make sure the RGB output levels below are set to limited.
BTW. without such drastic measures this option does nothing.

this can also cost the GPU more performance it's a complicated topic but worth a try because your GPU has similar bandwidth compared to a modern RX 6500 XT.

you could also try native DXVA2 but AMDs old decoder are really problematic and no hardware decoding could be better.

just to make that clear all of these recommendation are for such low end hardware and are for the most part not applicable to modern hardware.

ohh yeah don't waist your time on refreshrates you have other issues.

edit:
yes i almost forgot.
go to rendering -> dithering -> and set it to "none".

[dupcengier27]

kindly thanks everyone for valuable advice. In the section - Chroma upscaling - I chose "Lanczos", while in the sections - Image downscaling and Image Upscaling - DXVA2, when choosing Cubic or Lanczos there was a huge drop in frames ..... As acceleration in LAV Video I chose DXVA2 (copy-back ), the rest as in the attached screenshots.
https://i.ibb.co/bRXnybg/LAV-Filters-ustawienia-punkt-06-12.png
https://i.ibb.co/FzC70qq/Madshi-OSD-punkt-06-07.png
https://i.ibb.co/CwZHN6N/Madshi-ustawienia-chroma.png
https://i.ibb.co/Jj2XxZy/Madshi-ustawienia-image-downscaling.png
https://i.ibb.co/2Pk9KHg/Madshi-ustawienia-image-upscaling.png
https://i.ibb.co/Lz5Nns2/Pot-Player-OSD-punkt-06-06.png

[flossy_cake]

Your CPU is 2.7Ghz and Im guessing it should be able to decode 1080p up to 60fps, and maybe even 4k up to 30fps if it's not h.265. So another option is to turn off GPU decoding in LAV, that might free up enough GPU for MadVR stuff.

[huhn]

please don't assume hardware decoding will always cost GPU cycles they often even give you more to work with because you are usually trading very little processing for upload processing which is usually massively cheaper at least when compared to DXVa processing cost on most GPUs.

there are AMD card that will stop boost and such but i'm not aware if this is one of them.

btw. you should have other filter loaded which "shouldn't" be there like ffdshow video processor which could change the pixel format.

[dupcengier27]

Thanks again for any suggestions. However, due to the fact that after the last attempts with Madshi, there is no image playback, until I have a more powerful graphics card, unfortunately I decided to give up using Madshi in favor of MPC Video Renderer with the settings as in the screenshot, after the first attempts I did not notice a drop in frames. ... someday I will come back to the topic of Madshi, what would a sensible graphics card for Madshi have to be?
https://i.ibb.co/4TRk4N5/MPC-Video-Renderer-ustawienia.png

[James_b]

Hi guys!
Is there any recoder out there that can use madVR reconstruct sharp? I'm interrested in upscaling some DVD-clips before youtube.

[kasper93]

Hi guys!
Is there any recoder out there that can use madVR reconstruct sharp? I'm interrested in upscaling some DVD-clips before youtube.

I don't believe madVR is the right tool here. There are a lot solutions including ML based upscalers that will "restore" your DVD footage. Although needs experimenting what works best for your content and I personally would just upload native resolution.

Also please don't use reconstruct, it is too strong, introducing a lot of ringing and artifacts. Maybe if you have perfect luma... but it is rarely the case and for sure not for DVD...

Generally biliteral is my favorite [...]

I need to take it back... Biliteral Sharp can work good, but I tested it with gaming footage and it tends to destroy the details. Color outline around the edges, particle effects and similar details are mostly just "fit" to luma channel which tends to remove or reduce detail in chroma channel a lot. On the other hand if you have highly compressed chroma it can nicely clean out compression artifacts. It is double edged sword. It has to be used with caution and probably depending on source, it is not "universal".

After lots of testing I would recommend NGU Sharp/AA medium+ (low is bad) if you have gpu cycles to burn or for normal people Bicubic75AR is perfectly fine. Also DXVA can work nicely, but your milage may vary with divers/hardware.

[flossy_cake]

I need to take it back... Biliteral Sharp can work good, but I tested it with gaming footage and it tends to destroy the details. Color outline around the edges, particle effects and similar details are mostly just "fit" to luma channel which tends to remove or reduce detail in chroma channel a lot. On the other hand if you have highly compressed chroma it can nicely clean out compression artifacts. It is double edged sword. It has to be used with caution and probably depending on source, it is not "universal".

I am interested to hear your opinion on "bilateral old", if you wouldn't mind. My TV is not sufficient for evaluation as it downsamples to 4:2:2 internally.

[chros]

After lots of testing I would recommend NGU Sharp/AA medium+ (low is bad).
What do you mean? According to madshi there shouldn't be visible diff between low and med for chroma (I use Sharp).

[huhn]

can you share a sample where NGU AA low and med for chroma makes a difference?

and does super xbr not perform similar in these cases?

[mark0077]

Guys, do you NVidia owners know why the NVidia control panel setting called "Max Frame Rate" actually affects madVR performance, even if the max frame rate is set to something like 120fps. I notice even if playing say 23.976 content, although it doesn't cause frame drops, I notice some of the madVR buffers go down to say 0-1/8 or 1-2/8. I still would have thought its buffers would have filled up as the content is only 23.976fps. Is there any technical explanation for this, like does NVidia somehow limit software to 1 or 2 buffers when in this frame limit mode?

[flossy_cake]

Why not just disable it for the media player's exe in NVCP?

[mark0077]

Why not just disable it for the media player's exe in NVCP?

Oh yes that's what I have done. It was more out of curiosity to see if anyone understood technically why a 120fps limit causes 23.976 content to have the issue where the buffers don't fill up.

[flossy_cake]

Oh yes that's what I have done. It was more out of curiosity to see if anyone understood technically why a 120fps limit causes 23.976 content to have the issue where the buffers don't fill up.

Pure speculation, but maybe an optimal fps limiter limits the number of frames in the queue so that when it detects the fps is above the limit it doesn't have to wait for all the ones it the queue to be presented before the limiter actually comes into effect. Maybe it helps keep latency lower or something when fps is fluctuating near above/below the limit. I'm probably wrong but that's the best I can come up with.

Also if you use RTSS/Afterburner fps limiter, probably best to make a profile for the media player's exe to disable its fps limiter as well. Actually I disable it completely for mpc-hc64.exe as I don't want it injecting anything which might interfere with MadVR.

[x7007]

Does anyone have MadVR work with 22.4.2?? I can't get HDR passthrough with this driver. only Let Madvr Decide.

NVM: MadVR is buggy as hell, none stop bugs with AMD, I reset to default settings and installed back 0.92.7 instead 159, deleting from the registry wasn't enough

[LenniePaz]

Does anyone have MadVR work with 22.4.2?? I can't get HDR passthrough with this driver. only Let Madvr Decide.

NVM: MadVR is buggy as hell, none stop bugs with AMD, I reset to default settings and installed back 0.92.7 instead 159, deleting from the registry wasn't enough


Use v113. It doesn’t expire, and it’s stable.


Sent from my iPhone using Tapatalk

[James_b]

Hi Guys! madVR is a hq render, are there also hq decoders?

[Siso]

Use v113. It doesn’t expire, and it’s stable.


Sent from my iPhone using Tapatalk

What about the increased rendering times with 113, and the broken SM?

[LenniePaz]

What about the increased rendering times with 113, and the broken SM?


I’ve never had increased render times. They are the same on all versions. I have no use for SM.


Sent from my iPhone using Tapatalk

[huhn]

Hi Guys! madVR is a hq render, are there also hq decoders?

no. they are already bit perfect it can't get better.

[flossy_cake]

Hi Guys! madVR is a hq render, are there also hq decoders?

I guess that would be a function of the codec? I am using the k-lite codec pack which claims "The best available codecs are already selected for you by codec experts". But I'm not sure if those codecs are used when doing hardware accelerated decoding.

[x7007]

Use v113. It doesn’t expire, and it’s stable.


Sent from my iPhone using Tapatalk

Which version is that?
https://www.videohelp.com/software/madVR/old-versions#download

[x7007]

Use v113. It doesn’t expire, and it’s stable.


Sent from my iPhone using Tapatalk

Which version is that?


madVRhdrMeasure130.zip?

[LenniePaz]

Which version is that?


madVRhdrMeasure130.zip?


No, v113.


Sent from my iPhone using Tapatalk

[James_b]

Tapatalk

[x7007]

No, v113.


Sent from my iPhone using Tapatalk


Can you give me the link?? I can't find it

[Polopretress]

Just simple ==> replace XXX by the number of beta you want to download

http://madshi.net/madVRhdrMeasureXXX.zip

[x7007]

Just simple ==> replace XXX by the number of beta you want to download

http://madshi.net/madVRhdrMeasureXXX.zip

Thanks I got it now
What do you say fixed in this version that doesn't in all others?

[Polopretress]

I take this opportunity to ask a question about DTM that i like but on which i have a big issue of highlight with Sony projector serie x90. (european modele)
I have beta 157 and just switched to 159 but same issue, of course.

In fact, on some sequences (such as Aquaman) where highlights are very high, the highlights seen on this scene are very poor when DTM is active compared to Passthrough method.
I think i have tried everything to increase highlights when DTM is active but i have never reached the performance i can have in Passthrough mode.

It is like something , somewhere, is clipping the maximum IRE i can have when DTM is On.
The method i have found is to modify, in the tab "hdr", the value of "detail threshold" parameter to 10 (instead of OFF) to recover correct highlight when image has high IRE. But also, in addition, modify the setting of the projector on gamma 7 setting of Sony to have low gamma (1.8) on high IRE.

With these settings , it is acceptable, since these settings are only modifiying the brightness on image with high IRE (and keep unchanged the image on low/medium IRE)

does this problem has been already reported ?


below, the image i get without these strange settings (Gamma 7 and "detail threshold" = 10)


DTM OFF , Mode Passthrough (in HDR)
https://drive.google.com/file/d/1qwvyKfDa6CInejuT4H6AP-ujHmyr57jV/view?usp=sharing

DTM ON with contrast recovery at Low (in SDR)
https://drive.google.com/file/d/1luGI6Yqmak4XNtAPhE4UChHZOvSQ7_yz/view?usp=sharing

The problem with these pictures is that it does not reflect the real difference i can see on the screen because the highlights, even on the center are hugly higher with DTM OFF. (mode PassThrough HDR)