madVR - high quality video renderer (GPU assisted)

[andybkma]

Hi guys, is it a known bug that in mVR final version 92.17 (and in Beta 113 version at least) that after playing an HDR file and stopping the media player and then playing a non-HDR file all the colors are washed out? The only way for me then to play the non-HDR file correctly is to totally restart the media player. I can duplicate this every time with v113 and the last final 92.17 with both Zoom Player & Potplayer.

Reason I ask is because in Beta 165 this doesn't happen. Are there any settings that I can do please in final 92.17 or 113 to stop the washed out colors after playing an HDR file? I would rather just stick with final 92.17 if at all possible please...

[SamuriHL]

I vaguely remember that being an issue. It's possible madshi fixed that in one of the betas.

Sent from my SM-G998U1 using Tapatalk

[chros]

Yes, he did a workaround for it, you can take a look at the changelog (in my signature).

[mzso]

Where are test builds published by madshi?
He said he'll be releasing a new one, but I don't know where to watch for it.

[SamuriHL]

https://www.avsforum.com/threads/improving-madvr-hdr-to-sdr-mapping-for-projector-no-support-questions.2954506

It'll be posted in that thread. Eventually.

[SirMaster]

166 is here:

https://www.avsforum.com/threads/improving-madvr-hdr-to-sdr-mapping-for-projector-no-support-questions.2954506/page-799#post-62331962

[Celio1080p]

Hey there guys, sorry for this but..
What is the first thing i should try to improve quality ?
for example..
should i disable everything in
Trade Quality for Performance
?
I have a 3070 if that matters. Most videos i watch are 4/30 and 4/60
on a 4K TV Samsung, while a few are 1080/60
Windows 11 Pro.
Or it's better to not touch
Trade Quality for Performance
And change other settings first ?

[Msarc]

@Celio1080p

These are the "trade quality for performance" settings that were said to not actually impact quality to any visible degree, so you can leave them on:

https://i.imgur.com/XCMhowp.png

Overall, Image Upscaling makes the biggest impact on quality and is the main area you want to max out. You can even drop Chroma Upscaling down to 'Bilinear' and Dithering down to 'Ordered' if you need some performance back, without noticing any quality change beyond scrutinizing screenshots side-to-side.

[mzso]

166 is here:

https://www.avsforum.com/threads/improving-madvr-hdr-to-sdr-mapping-for-projector-no-support-questions.2954506/page-799#post-62331962

Great. Does it also include a nag message, that will cause trouble in a few months?

[Siso]

Great. Does it also include a nag message, that will cause trouble in a few months?

Naturally.

[huhn]

but that message may or may not crash this time!

[mzso]

Naturally.

Too bad. It would have been nice to have a recent build to rely on in the future, since the stable is ancient and no now one is in sight.

[mzso]

but that message may or may not crash this time!

Still a pointless nuisance. Especially knowing how rarely new builds are released these days. As a matter of fact, the crashing bug was needed to motivate madshi to release a new build.

[SamuriHL]

Given what madshi posted recently, it's not likely to ever happen that we get another non-time out madvr build. There was talk about adding a licensing mechanism but I wouldn't hold my breath on that ever happening.

[SamuriHL]

As a matter of fact, the crashing bug was needed to motivate madshi to release a new build.

Absolutely untrue.

[lvqcl]

Still a pointless nuisance.

From https://www.avsforum.com/posts/62306657/:
people need to understand that I can't spend development resources on the madVR HTPC software, if this comes back like a boomerang to hurt my business in a noticeable way. I think that should be obvious.
For example, one of our dealers demoed an Envy at some show recently. At the same show there was a commercially built and sold HTPC, running madVR, and not just as a source device, but also for video processing and HDR tone mapping purposes. Obviously our dealer was rather annoyed about that.

[mzso]

libplacebo is just a lib other renderer can use (including madVR).
it just depends what other make out of it.
if they add a profile system then yes but that's not the job of libplaebo which is just source code.

Any reason the MPC renderer guys are not using it?

[mzso]

From https://www.avsforum.com/posts/62306657/:

It would be nice if someone with sufficient right could update the OP with the current state of things, since madshi won't.

[huhn]

Still a pointless nuisance. Especially knowing how rarely new builds are released these days. As a matter of fact, the crashing bug was needed to motivate madshi to release a new build.

you could have just ask.
he even ask if he should do a version now or if it is ok to release it at the end of the week (which also didn't happen).
Any reason the MPC renderer guys are not using it?

ask them not me they are active in this forum.

[thighhighs]

Any reason the MPC renderer guys are not using it?I asked on a local forum. Short answer: Too much working time required. And they are not very interested.
Anyway, as huhn said, you can ask them yourself.

[jespermart]

Could someone explain how to use "scalingFactor.x/y" and also explain the use of "not" and "!"

A newbie in profiling:stupid:

[flossy_cake]

Could someone explain how to use "scalingFactor.x/y" and also explain the use of "not" and "!"


I've never actually used it but I'm pretty sure scalingfactor would be the amount that MadVR is scaling by, eg. if watching a 1920x1080 video and MadVR is upscaling it to 3840x2160, then scalingfactor.y would be 2 since 1080x2=2160.

The "not" and "!" would be used when evaluating a boolean eg.

if (not srcInterlaced) "my profile for non interlaced sources"


Supported boolean values:

4:2:0, 4:2:2, 4:4:4, RGB which pixel format does the source have?
srcInterlaced is the source interlaced?
filmMode is film mode (IVTC) active?
MPEG2, VC-1, h264 which codec is the source encoded in?
fseMode, overlay, windowed rendering mode
AMD, nVidia, Intel which GPU manufacturer are we rendering on?


I really wish MadVR had a way of setting the scalingfactor.x/y through a profile . That way we could control aspect ratio errors which are prevalent with SD content (there's this ambiguity about whether the width is 720 wide or 704 wide that isn't handled correctly by all mastering studios which creates sometimes a 2% discrepancy in the aspect). Currently I am having to do this with avisynth script because MadVR's screen config > anamorphic lens option cannot do it properly.

[mzso]

you could have just ask.
he even ask if he should do a version now or if it is ok to release it at the end of the week (which also didn't happen).

He last commented on this forum almost a year ago. Doesn't seem active.

[huhn]

you have to ask him on AVS

[huhn]

can someone else test nvidia DXVA scaler in madVR?

i get similar if not better results then jinc3 AR but only in madVR in EVR or mpcVR it is still looking like a bad bicubic implementation.
512.95 win 10 1060 6gb

i will update driver and do some comparison later.

[DVDIT]

I setup an older Dell T3610, PCIe 3.0 as HTPC for use with madVR and Video processor. My issue is the image quality of HDR is nothing like I see posted on forums. HDR looks just like SDR when playing thru madVR + VP. The image is dark and I don't see HDR color pop I see on my Sony LCD TV. I am not expecting a projector to match LCD in brightness but at least I expected good color pop good brightness given one of my screen is 2.8 gain. It must be that I have setup some options wrong but not sure where or what to look for and adjust. Is there a suggested setting for best picture quality?

My equipment:

JVC RS540 and Epson 5040UB both with under 450 hours on the bulb
Screen - Carada BW 1.1 gain and Da-Lite HP 2.8 gain
GPU RTX 3060 Ti OC

Cltr+J screenshot attached. Thank you

[ryrynz]

i get similar if not better results then jinc3 AR

Sounds promising. Looking forward to your comparison.

[huhn]

GT: https://abload.de/img/touhou-badapple.mp4_snkdod.png
output forced to NV12 to make DXVA not go crazy.
still 512.95
AR is used when possible.

bicubic 60: https://abload.de/img/bicubic60a2ern.png
DXVA2: https://abload.de/img/dxva2w6d8r.png
jinc: https://abload.de/img/jinc2yeo4.png
lanczos: https://abload.de/img/lanczos3mfcv0.png

the bad results:
NGU AA VHVH bicubic 60: https://abload.de/img/nguaavhvhbicubic605zf52.png
NGU SHARP VHx4: https://abload.de/img/ngusharpvhx4bicubic6090far.png
superXBR 100 150 bicubic 60: https://abload.de/img/superxbr100150bicubictley2.png

NGU sharp VH VH would have looked even worse.

this does not work on coloured images the DXVA results will be very soft in that case for some reason so kinda of a false positive.
while calling DXVA2 in this test better than jinc is a bit much thanks to some flaws. but i have no clue what's going on it is massively sharper with only some more ringing?
I setup an older Dell T3610, PCIe 3.0 as HTPC for use with madVR and Video processor. My issue is the image quality of HDR is nothing like I see posted on forums. HDR looks just like SDR when playing thru madVR + VP. The image is dark and I don't see HDR color pop I see on my Sony LCD TV. I am not expecting a projector to match LCD in brightness but at least I expected good color pop good brightness given one of my screen is 2.8 gain. It must be that I have setup some options wrong but not sure where or what to look for and adjust. Is there a suggested setting for best picture quality?

My equipment:

JVC RS540 and Epson 5040UB both with under 450 hours on the bulb
Screen - Carada BW 1.1 gain and Da-Lite HP 2.8 gain
GPU RTX 3060 Ti OC

Cltr+J screenshot attached. Thank you

fix the queues. try present in advanced 3 which is the absolute default recommendation anywhere.
type the correct peak brightness in madVR. you need to measure that. if you can't try a number between 50-100 nits. it's a PJ they don have brightness it will never look like a TV which is easily magnitudes brighter.

rules 12.

[Sunspark]

It's not impossible that Nvidia changed the internal algorithm they use for DXVA2 on their new cards. It would be interesting to see CPU/GPU power consumption to see if appears to be doing more work than expected.

[huhn]

it's a 1060. but this is not bicubic as far as i can tell.

GPU CPU is zero this is DXVA we are talking about

[Sunspark]

Nvidia is probably using bilinear.

[huhn]

does this look like bilinear to you? and no they never did always was some kind of bicubic in the past when working at least same for AMD and intel was lanczos with AR.

[Damien147]

I create a 3dlut for madvr with displaycal and then I only move the 3dlut file to a folder deleting everything else.The only needed file is the 3dlut file,right?

[huhn]

if you don't want to verify it yes.

[Damien147]

All good then,thank you.

[flossy_cake]

Silly me, I was using MadVR's file/folder tagging method of applying profiles instead of through the GUI, thinking it was the more convenient method. But I've now realised it definitely isn't, for at least 2 reasons which became apparent to me this evening:

1. As soon as you add a tag, the pathname of the media file changes, so if you are doing a backup of your [potentially massive] video library and you want to only merge new media files since the last backup, the newly added tag will cause it to be unique and thus create a duplicate on the backup drive (one with the new tag, one with the old or nonexistent tag) thus taking up twice as much space, extended copying times, and frustration of having to manually cull the duplicates afterwards.

2. Setting profiles through the GUI allows selection of more than 1 profile per media file, reducing the total number of profiles needed to get a desired outcome. For example suppose you have a sharpening profile and a zoom profile, you could link both of those to a single source file by placing eg. if (filepath == "*\movies\some movie\*") in BOTH profiles, and it will select both sharpening and zoom for that media file. Whereas with tags you can only choose a single profile to apply, i.e you can't select 2 profiles by tagging something like "c:\movies\some movie [profile=sharpen][profile=zoom]". With tagging this means you'd need to create an extra 2 profiles with a single unique name just for that media file to get the same result, thus resulting in a lot of excess profiles in the long run. Whereas profiling through the GUI everything is modular and extensible.

Question tho: if I backup MadVR's settings.bin, does that save everything in the GUI including profiles, and will it work with other copies/versions of MadVR on other machines? I want to be absolutely sure I'm backing up everything as many hours have been invested in my configuration settings and losing it would be devastating.

:thanks:

[QBhd]

Yes the settings.bin backs up everything. I never understood why people tag files for profiles. I think there may be some use for tags (even though I have never used it), but for profiling the tools in the GUI are so powerful.

QB

[brazen1]

Silly me, I was using MadVR's file/folder tagging method of applying profiles instead of through the GUI, thinking it was the more convenient method. But I've now realised it definitely isn't, for at least 2 reasons which became apparent to me this evening:

1. As soon as you add a tag, the pathname of the media file changes, so if you are doing a backup of your [potentially massive] video library and you want to only merge new media files since the last backup, the newly added tag will cause it to be unique and thus create a duplicate on the backup drive (one with the new tag, one with the old or nonexistent tag) thus taking up twice as much space, extended copying times, and frustration of having to manually cull the duplicates afterwards.

Most of my titles are tagged before ever being added to backup. For a title already in backup, I rename it before backup thinks there's a new title to add.

2. Setting profiles through the GUI allows selection of more than 1 profile per media file, reducing the total number of profiles needed to get a desired outcome. For example suppose you have a sharpening profile and a zoom profile, you could link both of those to a single source file by placing eg. if (filepath == "*\movies\some movie\*") in BOTH profiles, and it will select both sharpening and zoom for that media file. Whereas with tags you can only choose a single profile to apply, i.e you can't select 2 profiles by tagging something like "c:\movies\some movie [profile=sharpen][profile=zoom]". With tagging this means you'd need to create an extra 2 profiles with a single unique name just for that media file to get the same result, thus resulting in a lot of excess profiles in the long run. Whereas profiling through the GUI everything is modular and extensible.

What's the problem with lots of profiles? I'd rather have lots of unique profiles using tagging for unique titles than have a bunch of special folder sources for unique titles specifically. That way, I don't have to 'sort' any titles to where they belong. I let madVR sort the titles.

For me, tagging can pick one specific title and apply unique madVR functions without affecting any other titles. I wouldn't want that title to need its own unique location or I'd end up with too many file/folder paths that would be confusing.

For example. I have six bizarre titles. 3 of them need 100% more saturation. The other 3 need 25% more saturation. I'm not creating 2 new path sources for 6 titles. Instead, I edit the file names to include tags. *SAT100* to 3 of them and *SAT25* to the other 3. Now these titles will live amongst every other title but madVR knows to treat them differently.

When I playback one of these titles, the extra profile for it via tagging apples extra saturation that no other titles get, unless they have the unique tag too.

At the same time, a saturation needy title shares the same metadata profile as other titles. So, not only does it use its unique tag adjustment in madVR, it also gets other profiles applied to it that are shared by other titles that are not saturation needy.

For example. Maybe its 1080p. All 1080p titles have unique profiles that are different from say 2160p. So not only is the extra saturation applied, so is the 1080p profile. Since not all 1080p is created equally, these are separated into even more profiles. 1080p SDR vs 1080p HDR for example. So, tags are not limited to only one profile.

Multiple profiles are applied to one title with or without tagging. This is because every tab in madVR can contain (its) own profile too. Each tab can have a profile group which also contains multiple profiles.

Imo, this is a huge beauty exclusive to madVR. It gives me the ability to ultra fine tune titles without having to assign them special addresses separate from the rest of the community. And playback of all titles is automated perfectly with madVR doing all of the guess work effortlessly and quickly.

Lastly, there are times I want to know exactly what I'm watching. I use extensive file names so I can quickly understand exactly what it is and Kodi info or MediaInfo are not going to let me know I'm applying extra saturation using madVR unless I read the file name with that SAT25 tag.

That said, we all do some things differently because it works best for our environment so keep on keepin' on :)

[flossy_cake]

Most of my titles are tagged before ever being added to backup. For a title already in backup, I rename it before backup thinks there's a new title to add.

Well, currently I am working my way through seasons of DVDs one at a time, and I find early seasons tend to look poorer quality so I'm making an adjustment at the end of each season. By the next time I do a backup there could be lots of new tags and I would have to keep a running list of these so I know which duplicates to delete from my merged backup due to the new filenames.

But some of the adjustments like gamut, matrix & black level can only be done through a tag, so I will still have duplicates in my merged backups :(

But another advantage to GUI profiles is that say you want to apply a profile to a season with 22 episodes and 1 episode is different and needs different profile correction. With tags you can't tag the top level folder and then make an exclusion tag for that 1 episode because the top-level folder tag overrides it, whereas in GUI you could exclude it with something like and (filepath != "*\series\show name\season 1\S01E01*")



Lastly, there are times I want to know exactly what I'm watching. I use extensive file names so I can quickly understand exactly what it is and Kodi info or MediaInfo are not going to let me know I'm applying extra saturation using madVR unless I read the file name with that SAT25 tag.

This is actually a very good point that I did not consider - in Kodi I can see the filename has a tag before I choose to play it so I know what I'm getting. Well, this will be a disappointing loss for me. It would be nice if MadVR put a little overlay for the first 10 seconds of playback to let you know a profile is active.

Well, now I have to reconsider... I'm going to do my first large backup tomorrow which will take 14 hours so I want to get it right the first time without messing everything up.

edit: perhaps what I'll do is continue to use tags and use an Autohotkey script to identify duplicates to delete after a merged backup.

[brazen1]

By the next time I do a backup there could be lots of new tags and I would have to keep a running list of these so I know which duplicates to delete from my merged backup due to the new filenames.

I do. I have a folder especially for anything that is destined for my weekly backup. In that folder are links to what I need to copy over. Those links are usually just .txt files and I name them to what needs attention. So I name a .txt file the new renamed title because it got a new tag. Then when backup is open, I just rename the file. No need to delete the whole thing and readd it just for a name change. Not sure how you backup but surely you can edit file names?

But some of the adjustments like gamut, matrix & black level can only be done through a tag, so I will still have duplicates in my merged backups :(

And/Or, a profile name added to a profile group.

But another advantage to GUI profiles is that say you want to apply a profile to a season with 22 episodes and 1 episode is different and needs different profile correction. With tags you can't tag the top level folder and then make an exclusion tag for that 1 episode because the top-level folder tag overrides it, whereas in GUI you could exclude it with something like and (filepath != "*\series\show name\season 1\S01E01*")

Not so sure about that? I think the moment you re-tag a tagged item within a tagged season, it gets plucked from the herd because it's going to get its own profile. Then you adjust its profile just like the rest of the herd in addition to whatever it was you changed making it different. But, you'd have to test to be sure.

It would be nice if MadVR put a little overlay for the first 10 seconds of playback to let you know a profile is active.

Or, if you use nfo's with Kodi, you can edit the 'Tag' line to read whatever you want. The Tag line always displays 1st in the synopsis. It's even colored in bold. It won't show you active madVR profiles but it will show something like "25% Xtra Saturated SDR with enabled WindTrack - Set Bass Shaker volume to 75%". And then the rest of the synopsis follows by auto scrolling.

[flossy_cake]

you can't select 2 profiles by tagging something like "c:\movies\some movie [profile=sharpen] [profile=zoom]"

Just realised this functionality can be added with something like if (filepath == "*[profile=thisProfile]*") "thisProfile"

Then to identify duplicates after a merged backup, something like...


for each file in video library
if FileExist(StripMadvrTags(thisFile))
Duplicates.push(thisFile)

But I'm also renaming files/folders for other reasons besides profile tagging, which would still create duplicates after a merged backup. eg. I recently prefixed every season of a show with the year it was released to ensure chronological ordering in file explorers - that entire series would get duplicated at my next merge. I'm also often renaming folders when adding Avisynth scripts (eg. splitting a folder into a "source" subfolder and an "avisynth" subfolder containing just .avs script files mirroring the folder structure of source). Not sure how to identify these duplicates without involving large amounts of CPU processing :confused:



But some of the adjustments like gamut, matrix & black level can only be done through a tag
And/Or, a profile name added to a profile group.


How do you mean? I can't find any MadVR GUI settings that correspond to the [matrix], [primaries] and [blacklevel] tags.

[flossy_cake]

Just realised this functionality can be added with something like if (filepath == "*[profile=thisProfile]*") "thisProfile"


I suppose this could also be used to allow child exclusions when tagging parent folders, eg to make an exclusion for S01E09 to force sharpen high just for that one episode:

C:\Some TV Series\Season 1 [profile='sharpen low']\S01E09 [profile='sharpen high'].mkv

In this case by default MadVR will still apply sharpen low to S01E09 since it's a parent folder which overrides the child tag for S01E09.

To workaround this we might put in sharpen low profile:

and (filepath != "*[profile='sharpen low']\*[profile=*")

This should disable sharpen low since it found another profile tag in a child folder.

I haven't tested any of this yet.

[brazen1]

How do you mean? I can't find any MadVR GUI settings that correspond to the [matrix], [primaries] and [blacklevel] tags.

Oh, you're right. That must be tonemapping. I don't use them. Not familiar.

[flossy_cake]

By the next time I do a backup there could be lots of new tags and I would have to keep a running list of these so I know which duplicates to delete from my merged backup due to the new filenames.I do. I have a folder especially for anything that is destined for my weekly backup. In that folder are links to what I need to copy over.

I think I'll do it this way too - by keeping a running note of what folder names I changed with tags or other things. Because it occurred to me that if for example I added a tag to a 150GB series folder, I would have to copy that whole 150GB folder duplicate and then delete the original after the merge, wasting time and space (think of what happens when I start getting near drive capacity, I'd need to have 150GB of storage headroom just to temporarily accommodate this duplicate) when instead I could just consult my list before starting the merge.

[flossy_cake]

I haven't tested any of this yet.

I tested it and it works, but only if I use my own tagging nomenclature otherwise MadVR overrides it, eg. using [pro=sharpen] instead of [profile=sharpen]. Then I can gain functionality of: 1. applying multiple profiles simultaneously with multiple tags, and 2. making it so child tags override parent tags (useful for specifying exclusions to a top-level folder tag).

Solution:


if (filepath == "*[pro=sharpen]*") and (filepath != "*[pro=sharpen]\*[pro=*") "sharpen"



But, there is still a limitation where you can't apply multiple profiles from the same group, eg. [pro=sharpen][pro=sharpen2] wont apply both. But I can't think of a good reason why this functionality would be needed. Whereas [pro=sharpen][pro=denoise] does work to activate both because denoise is in a separate profile group to sharpen.

[brazen1]

I think the profiles and settings used with or without tags should only be 'universal' and not all processing should be universal. By universal I mean, videos or groups of videos that are hard set with settings - they are always going to have certain settings applied because they are permanently set in madVR. Assume these videos will be played on different displays still processed by madVR and whatever else you might be using. Some settings might be great on one display... but not another. So, those types of settings wouldn't be universal.

But we still want to use those settings when appropriate. So, it might be best to keymap some settings to your remote control for on-the-fly adjustments. For instance. You might tag a denoise setting into a video file name. It's going to use that setting universally. Then when you playback on a different screen, it might not look right. So, use the denoise keymap to readjust it. Using a keymap, some are simple toggle on/off by pressing the keymap twice. Some have values like 1-10 such as denoise. These require two keymaps. One for values up and one for values down. Values down also turns the process off once you reach zero.

I like that the OSD for on-the-fly settings is small but large enough to see it. It's not intrusive and guests will never notice when you make a quick adjustment(s). Same goes for the clock OSD etc.

Just thinking you might not want to universal code, especially tag, some things that might be better off keymapped. Depends on your environment and usage. Just sayin'

[Sunspark]

I use D3D9 Overlay.

I wonder why sometimes when starting a video render times will be 2.5-3ms and other times the same video it will be 15-20ms. There doesn't seem to be any visible frame drops, stutters or repeats but I think there is more power being consumed. I cannot consistently make it be one or the other rendering time, it appears random, but I don't know why. It happens during a resolution/refresh rate change from native desktop to a lower resolution/refresh.

Has anyone observed this behaviour before?

I did some experimentation, lowering GPU buffer to 4 from 8 wasn't the solution so I put it back to 8, even though this is an igpu and doesn't have dedicated vram. I had backbuffers at 1, so I tried putting it back to the default 3. Wasn't the solution either. Backbuffers 4 didn't change anything either. Hmm.

[huhn]

just different GPU power state.

[Sunspark]

How do you figure power state? It doesn't change during playback that I saw.

I did observe that gpu queue at 4 was an improvement because it works to slice 5 ms off when the glitch happens, but I haven't been able to eliminate it entirely. Currently I am experimenting with cpu queue 12, gpu queue 4, backbuffers 4. Before it was 16/8/1.

[Celio1080p]

Hey there. Thank you very much xD

Guys, huh.. what is the best option to increase sharpeness of the videos ? mainly 4k videos.
Rarely 1080p.
https://i.imgur.com/ZTKgYRG.jpg
I always used the Sharpen Shader inside MPC-HC (Long time ago)..
And now i'm using the Sharpen Shader inside MPC-BE.. it's the same one, probably.

Anyway.. i suppose i'm not doing this the right way ?
Or i am ?

See the screenshot. For me, the quality is pretty good. I'm not complaing.
But perhaps..
Maybe it would better to turn it off, and increase it by using MADVR ?
I always liked sharpened picture. I hate blur pictures..
Just wanted some opinions xD