madVR - high quality video renderer (GPU assisted)

[huhn]

wow...
film mode is madVR only and needs a copyback decoder and by default it looks only at the centre of the frame the GPU driver is not related to that at all. and this is not deinterlacing this is detelcine.

the deinterlacer needs a couple of frames before it works with NVidia intel or AMD.
and why it needs double the processing power is again very obviously.
madVR now does double the amount of processing because it has to process double the number of frames.

[flossy_cake]

wow...film mode is madVR only

Oh ok, I was going based on what you wrote here:

film mode is IVTC.
madVR doesn't have a deinterlacer it just used the GPU provided deinterlacer if that's bad you get bad results if it is good you get good results. a good GPU driver deinterlacer is currently the best available deint we have for real time usage.

this is not deinterlacing this is detelcine.

Well it is a conversion from 29.97i to 23.976p



and why it needs double the processing power is again very obviously.
madVR now does double the amount of processing because it has to process double the number of frames.

However much processing power it takes, it's still bugger all so something is wrong with his setup as it shouldn't be eating 80% GPU and making fans spin up. Even on my R9 380 it's only 14% GPU usage at 850mhz clock.

[flossy_cake]

So just to clarify:

1. [deint=film] and [deint=ivtc] do exactly the same thing, which is to enable MadVR's implementation of deinterlacing/detelecining for interlaced video with film cadences like 3:2, 2:3, 2:2 and maybe even some other fancy ones used by animation, but NOT 1:1 cadence (ctrl+j reports 'unknown cadence' on that F1 clip and playback is very stuttery).

2. [deint=video] hands the deinterlacing off to the GPU driver, and the quality of deinterlacing you get is dependent on the GPU driver.

Is my understanding correct?

[huhn]

1. just an alias and no this operation is detelecine not stricked deinterlace.
or call it inverse telecine. if you like that term.

60i to 23p is detelecine. 60i to 60p is deinterlacing.

it is quite limited.
if 23p is available it will use it for 60i source so a 2:2 cadence will not work properly in that case. if you have a 2:2:2:4 source it will only properly work with 23p output mode not at 120 hz.
2:2 works properly at 60 HZ. it kinda odd.

2. yes the deinterlacer is part of the GPU driver and depends on hardware and driver version.
madVR uses DXVA2 for that.

the called function is always deinterlacing even when detelecine (techincally this is only field matching not detelecine) is applied by the GPU driver the output is still 60p.

i ignore 50i here but you are very save with 50 hz output and video mode in general it just cost double the processing power.

there are user that kill 3080 with 50 or 60p/i source having high rendertimes is not uncommon.

[flossy_cake]

1. just an alias and no this operation is detelecine not stricked deinterlace.
or call it inverse telecine. if you like that term.

I know what you mean I only used that word cause madvr puts that option under deinterlacing category & tag [deint] so I was just referring to it as such but whatever it's just a pointer to an idea, we both know what the idea is, doesn't really matter what word we use as the pointer to the idea.

It's cool how madvr will even output 24hz over HDMI when [deint=film/ivtc] for 29.97i content -- good for DVD remux of some rare titles.


there are user that kill 3080 with 50 or 60p/i source having high rendertimes is not uncommon.

Well I have no idea what's going on like I said only 7% and 14% GPU usage on 1070 and R9 380 for that 1080i 1:1 clip. The codec shouldn't have anything to do with it but it's h.264 4.5mbps taken from local TV broadcast.

edit: rendertime increases from 6.5ms to 8.5ms for that 1:1 clip with [deint=off] vs [deint=video].

But I just remembered chroma alignment is broken when using dxva deinterlacing isn't it?

edit: ok when I untick "use DXVA chroma when doing DXVA deinterlacing" , render time increases to 10ms total, but GPU usage still the same. Can't really complain with +4ms for deint of 1080i.

edit: nope! The rendertimes on ctrl+J are still a function of GPU clockspeed. With max GPU clock it's only 4ms total render time with 1080i 1:1.

edit: finally with max GPU clock and untick "DXVA chroma when DXVA deint" I'm seeing huhns 2x increase in GPU usage for deint: 4% increases to 9%. Still effortless tho.

[el Filou]

Are you also using NGU medium upscale then downscale like fourbigkids is?
If their GPU is already at 34% for 25 fps it's understandable it's at 70% for 50 fps, the question is why the 34% at 25 fps.
Maybe they're using a heavier madVR config than you, or the GPU is downlocked, I don't know.both AMD and Nvidia seem to think sections of the red and white curbs at the bottom are weaveableIt's not so much that they 'think', rather than their algo is not smart enough to detect the motion so it gives up. Circuit curbs is a really difficult thing to deinterlace as it's very difficult to guess which way the motion is going, I see it also on onboard cameras depending on the speed of the car, it gives a pink blob mix of the two colours.

[fourbigkids]

Just back from a trip. Thanks for the continuing interest in this issue guys.

I have looked at my player settings. I am using MPC-BE with LAV filters.

LAV Video Decoder, Software Deinterlacing set at YADIF 50p/60p (Video); Madvr deinterlacing set to 'automatically on if needed': GPU 85degrees and 70%, video in Madvr stats identified as 50fps.

LAV Video Decoder Software Deinterlacing set to: Off; Madvr deinterlacing set to 'automatically on if needed': Temp and performance and reported fps remained the same.

LAV Video Decoder, Software Deinterlacing set as YADIF at 25p/30p (Movie): performance and temp dropped to 34% and 50 degrees respectively with 25fps reported in Madvr.

Disabling deinterlacing in LAV Video Decoder AND Madvr: performance and temp dropped to 34% and 50 degrees respectively with 25fps reported in Madvr.

I will check further, but noticed nothing significant in terms of video quality.

I will now try to get my head around the recent posts. Thanks again.

[fourbigkids]

Are you also using NGU medium upscale then downscale like fourbigkids is?
If their GPU is already at 34% for 25 fps it's understandable it's at 70% for 50 fps, the question is why the 34% at 25 fps.
Maybe they're using a heavier madVR config than you, or the GPU is downlocked, I don't know.

Hi el Filou. Are you suggesting I might need to lighten the load on madvr? Would it help pinpoint the problem if I posted images of some of the settings? If so, which would be the more power hungry so I don't flood the thread with more images? Cheers.

[flossy_cake]

Are you also using NGU medium upscale then downscale like fourbigkids is?

My display isn't > 1080 so I can't do NGU upscale, only NGU chroma, which if enabled increases my GPU usage to 20% from my baseline of 10% with bilateral chroma (max clocks). His 2060 is about 10% faster than my 1070, so with 2060 doing NGU upscaling to 4k, 35% GPU usage seems like a realistic figure.

or the GPU is downlocked

I thought so but his temps are high as well. I speculated it could just be the fan curve is set to allow higher temps in an attempt to reduce fan noise, but I'm doubting that as well as it seems related to deinterlacing settings somehow.

Disabling deinterlacing in LAV Video Decoder AND Madvr: performance and temp dropped to 34% and 50 degrees respectively with 25fps reported in Madvr.

Ok so definitely a deinterlacing issue. The "software" deinterlacing in LAV is done on CPU presumably so it shouldn't be increasing GPU temps.

Seems to be an issue with MadVR's deint settings. Could you please leave deint disabled in both LAV & MadVR & check that F1 clip I posted earlier which is already tagged with optimal MadVR deint setting.

I also see LAV has another setting in the Settings groupbox (auto/aggressive/force/disabled) but I have no idea what that actually does. I can't get it to produce any visible difference in any of the clips.

Also some settings require a restart to take effect.

[fourbigkids]

....

Ok so definitely a deinterlacing issue. The "software" deinterlacing in LAV is done on CPU presumably so it shouldn't be increasing GPU temps.

Seems to be an issue with MadVR's deint settings. Could you please leave deint disabled in both LAV & MadVR & check that F1 clip I posted earlier which is already tagged with optimal MadVR deint setting.

Also some settings require a restart to take effect.

Hi flossy_cake. Yes, I have been restarting after each setting change. I disabled deint in both LAV Video Decoder and in Madvr. Played your video and GPU processing went to around 76% and temp mid eighties when fan kicks in and it hovers around there.

[flossy_cake]

Hi flossy_cake. Yes, I have been restarting after each setting change. I disabled deint in both LAV Video Decoder and in Madvr. Played your video and GPU processing went to around 76% and temp mid eighties when fan kicks in and it hovers around there.

Ok so according to huhn, that F1 clip with [deint=video] tag will activate deinterlacing in the GPU driver, so the high usage and temps would seem to be an issue with GPU driver.

To confirm this could you check the other clip (movie with 2-2 cadence) as that one is tagged with [deint=film] and will use MadVR deint instead of GPU driver deint. Then if load & temps are good we know it's a GPU driver issue.

[fourbigkids]

Ok so according to huhn, that F1 clip with [deint=video] tag will activate deinterlacing in the GPU driver, so the high usage and temps would seem to be an issue with GPU driver.

To confirm this could you check the other clip (movie with 2-2 cadence) as that one is tagged with [deint=film] and will use MadVR deint instead of GPU driver deint. Then if load & temps are good we know it's a GPU driver issue.

Thanks for persisting with this flossy_cake, beyond the call etc... Have now done this and results are much lower. Performance sitting around 36% and temp low 50s. But not the much lower results I get with the less GPU intensive VLC or an alternative renderer.

So you are thinking a driver issue. Since noticing the issue I have updated the driver at least twice in the hope it might be resolved. (I use the studio version for reliability). Should I look at an older driver, or, try and reduce the load imposed by the settings I have in madvr processing and scaling? What do you think? Cheers.

[flossy_cake]

Thanks for persisting with this flossy_cake, beyond the call etc... Have now done this and results are much lower. Performance sitting around 36% and temp low 50s. But not the much lower results I get with the less GPU intensive VLC or an alternative renderer.

So you are thinking a driver issue. Since noticing the issue I have updated the driver at least twice in the hope it might be resolved. (I use the studio version for reliability). Should I look at an older driver, or, try and reduce the load imposed by the settings I have in madvr processing and scaling? What do you think? Cheers.

Looking at Nvidia's description it seems the studio driver should be better at deinterlacing as it's for video creators -- maybe that's why it uses more GPU cause it's using a better deint algo? I guess you could try the normal driver version and see if it's any different.

In any case we've isolated the GPU deint algo as the main culprit (since avoiding it appears to have reduced usage from 76% to 35%) so you can set MadVR to "force film mode" and you'll avoid the issue without having to tag files. But you wont be able to play interlaced 1:1 cadence files like that F1 clip, but those files are rare.

NGU is expensive so try cheaper ones like Cubic, Lanczos, bilateral or super-xbr. I'd anticipate at least a 10% reduction in GPU usage, but really you gotta check your clockspeed otherwise that value is meaningless. Or just set mpc-be64.exe to max clock in NVCP so you know what the "true" usage is.

I use Lanczos for upscaling/downscaling and bilateral old for chroma upscaling. I don't really see a benefit of the expensive ones, but perhaps you would on a larger 4k display.

It seems the scaling algos take effect immediately when you hit Apply in MadVR settings so you can just pause on a scene and preview them.

[fourbigkids]

Thanks flossy_cake. I'll get on with that further testing and report back. Do I need to turn madvr's deint back when I select force film mode?

[flossy_cake]

Do I need to turn madvr's deint back when I select force film mode?

Assuming you don't want the hassle of manually tagging files then I think the setting you'd want is:


☑ automatically activate deinterlacing when needed
◎ if in doubt, activate deinterlacing
◉ if in doubt, deactivate deinterlacing

☑ disable automatic source type detection
◉ force film mode
◎ force video mode

☐ only look at pixels in the frame center


If you do tag a file with [deint=film/video] I believe it will override all of the above settings except for "only look at pixels in frame center" .

[fourbigkids]

Thanks flossy_cake. I appreciate the help. As a token I would like to have named my first-born after you, but I have already named her 'Asmodian', although I am not sure she is completely happy with it. :-) Second born perhaps? :-) Cheers.

[huhn]

Looking at Nvidia's description it seems the studio driver should be better at deinterlacing as it's for video creators -- maybe that's why it uses more GPU cause it's using a better deint algo? I guess you could try the normal driver version and see if it's any different.
please stop spreading misinformation the deint algorithm is completely "free" like all DXVA2 processing. it's the doubling of the frame rate that cost processing power.
and no it's not better on studio driver because no one touches this and a professional is the last person to use it anyway.

In any case we've isolated the GPU deint algo as the main culprit (since avoiding it appears to have reduced usage from 76% to 35%) so you can set MadVR to "force film mode" and you'll avoid the issue without having to tag files. But you wont be able to play interlaced 1:1 cadence files like that F1 clip, but those files are rare.
no you didn't he was using software deint before... yadif.

[flossy_cake]

please stop spreading misinformation the deint algorithm is completely "free" like all DXVA2 processing. it's the doubling of the frame rate that cost processing power.

Yes but for some reason on his system the DXVA deinterlacing is causing high GPU usage.


no you didn't he was using software deint before... yadif.

Yes, and he disabled all deinterlacing and GPU usage reduced by half, then we tried DXVA deint via [deint=video] and GPU usage doubled, then we tried MadVR deint via [deint=film] and GPU usage halved.

I'm am cautiously optimistic that if he avoids DXVA deint and sets scaling algos to something less intensive than NGU, GPU usage should be around the 15-25% range at max clock.

[flossy_cake]

Ok I just observed on my system with 1080i 1:1 cadence [deint=video], ctrl+J is saying 25.000fps whereas on his it's saying 50.000fps.

So yeah maybe it is the double frame rate after all.

edit: yep setting YADIF deint causes ctrl+J to report 50.000fps and GPU usage doubles.

So yes it's probably not DXVA's fault after all. Probably the NGU upscaling to 4k @ 50fps.

edit: but then why does he still get high GPU usage when YADIF is disabled:


LAV Video Decoder, Software Deinterlacing set at YADIF 50p/60p (Video); Madvr deinterlacing set to 'automatically on if needed': GPU 85degrees and 70%, video in Madvr stats identified as 50fps.

LAV Video Decoder Software Deinterlacing set to: Off; Madvr deinterlacing set to 'automatically on if needed': Temp and performance and reported fps remained the same.

[huhn]

come on that's not the point.

you just say if we double the frame rate we have double the GPU usages. yes you have...

but film mode will not work on interlaced sources only telecine or PsF.

don't recommend to disable deint as soon as you finds a file that is interlaced he is screwed because of bad settings.

just use the profile and live a happy life.

[flossy_cake]

you just say if we double the frame rate we have double the GPU usages. yes you have...

Yes I acknowledge that, sorry if it was a late edit which you might not have seen:


edit: finally with max GPU clock and untick "DXVA chroma when DXVA deint" I'm seeing huhns 2x increase in GPU usage for deint: 4% increases to 9%.

So I think he should be fine with the following settings:

- Not NGU scaling
- Disable software deint in LAV decoder

As for optimal MadVR deint settings, unfortunately that is not really possible as MadVR's "automatic source type detection" doesn't work:

auto detection doesn't work.I just checked it again now and you're right -- it doesn't work. It seems to just use video mode deinterlacing the whole time regardless of the source cadence.

...so we have to set it manually to either film or video mode depending on the the content. Either by using tags or manually going into the settings dialogue.

[fourbigkids]

....

just use the profile and live a happy life.

Hi huhn. Just so I am clear, are you suggesting two profiles, one for high deintfps and another for low deintfps with the high one having less intensive madvr processing? El Filou has previously posted the following script:

if deintFps < 31 "LFR"
else "HFR"

When creating a profile madvr asks what processing, scaling algorithm or rendering setting is to be profiled. Can you suggest which of these that I should focus on reducing to achieve best tradeoff in quality vs performance?

Cheers,

Allan

[inzomgnia]

Hi all, looking for some help with something that may be a bit odd compared to the norm but hoping someone can point me in the right direction.

I'm using MPC HC + MadVR to upscale my capture card feed. The issue that it seems aliasing is only being applied to black/white edges and not colour edges. I can't figure out what's wrong.

I've seen somebody else do this so I'm pretty sure it's possible but I've spent quite a long time going through MadVR with no luck. I can confirm we are both using RGB24 inputs so there wouldn't (or shouldn't) be any initial chroma upscaling.

Here is what I found online:
https://screenshotcomparison.com/images/1608006057_8206315497.png
and here is what I have:
https://i.imgur.com/BXh6jjK.png

Anything grey/black/white is pretty much perfect, but anything with colour is a jaggy mess. It's especially noticeable on the red "Online" icon in the bottom left. Any advice would be much appreciate!

[Siso]

If I use beta 113 and then update to the latest betas which expires after time, and return to beta 113, will I be able to use 113 without expiration again?

[huhn]

Hi huhn. Just so I am clear, are you suggesting two profiles, one for high deintfps and another for low deintfps with the high one having less intensive madvr processing? El Filou has previously posted the following script:
if deintFps < 31 "LFR"
else "HFR"

yes.
that was an excellent advise.

When creating a profile madvr asks what processing, scaling algorithm or rendering setting is to be profiled. Can you suggest which of these that I should focus on reducing to achieve best tradeoff in quality vs performance?

Cheers,

Allan

depends all on your what you like too see in the end.
scaler can be very expensive so can error diffusion dithering options.
artifact removal can also be very expensive. i don't know waht you are using and even if i would know it still don't know what you want.

Hi all, looking for some help with something that may be a bit odd compared to the norm but hoping someone can point me in the right direction.

I'm using MPC HC + MadVR to upscale my capture card feed. The issue that it seems aliasing is only being applied to black/white edges and not colour edges. I can't figure out what's wrong.

I've seen somebody else do this so I'm pretty sure it's possible but I've spent quite a long time going through MadVR with no luck. I can confirm we are both using RGB24 inputs so there wouldn't (or shouldn't) be any initial chroma upscaling.

Here is what I found online:
https://screenshotcomparison.com/images/1608006057_8206315497.png
and here is what I have:
https://i.imgur.com/BXh6jjK.png

Anything grey/black/white is pretty much perfect, but anything with colour is a jaggy mess. It's especially noticeable on the red "Online" icon in the bottom left. Any advice would be much appreciate!

something scales the chroma up and something maybe even downscales it.
nothing can be done here about that in the madVR thread.
If I use beta 113 and then update to the latest betas which expires after time, and return to beta 113, will I be able to use 113 without expiration again?
yes why would it be any other way? 113 doesn't have the code so how could it run it.
beware 113 is buggy.

[Siso]

beware 113 is buggy.


Because of smooth motion?

[huhn]

yes that's definitely broken but there where other issues.

[fourbigkids]

I have created profiles for where 25p is using NGU image and chroma upscaling and 50i is using Spline for both. Performance is around 45% and temp is about 70degrees. Problem resolved, although the GPU does seem to be still working hard. I'll keep an eye on it.

I am using a Gigabyte GeForce RTX-2060-Windforce 6G OC Card. Could anyone recommend a card of similar size or smaller that would find the going a little easier while remaining quiet?

Thanks all for the help.

Cheers,

Allan

[chros]

It's not about the card: you have to downclock it with msi afterburner (limit its top clock speed).
eg my 1060 is limited to 1569 MHz @ 0.800 V, so it won't get hot (~62C) and the fan speed is low (¬750rpm max).
you can use monitor of nvidiainspector to check how your card operates during usage.
about creating complex profiles (https://www.avsforum.com/threads/guide-building-a-4k-htpc-for-madvr.2364113/post-60096583)

[Siso]

yes that's definitely broken but there where other issues.

So which beta do you recommend?

[Balling]

If I use beta 113 and then update to the latest betas which expires after time, and return to beta 113, will I be able to use 113 without expiration again?

Cannot you just launch through an app run as date, that owerwrites date and time API.

[flossy_cake]

I'm using MPC HC + MadVR to upscale my capture card feed. The issue that it seems aliasing is only being applied to black/white edges and not colour edges. I can't figure out what's wrong.
https://i.imgur.com/BXh6jjK.png


Something in your capture card setup is downscaling the chroma to what looks like quarter resolution (4:2:0)?

Normally 4:2:0 doesn't look that bad if the content is sourced from a camera which is already slightly blurred/anti-aliased before the chroma is downscaled. But this isn't the case with CGI which is a lot sharper -- that's why 4:2:0 is generally considered unacceptable for PC desktop use but ok for video content.

In my experience NGU sharp chroma upscaling really exacerbates chroma artefacts -- I'd recommend "bilateral old" for smoothing it out to something acceptable.

[huhn]

So which beta do you recommend?

i don't know 112.
newest and pray it doesn't fall apart.

[Siso]

i don't know 112.
newest and pray it doesn't fall apart.

Thanks, huhn. Does anyone have a link to 112? Can't find it on the net.

[el Filou]

Random observation but maybe it applies to more recent GPUs too and could be useful to someone one day: I've noticed testing flossy_cake's interlacing clip on my desktop that on my old Radeon HD 7870, deinterlacing stops working correctly with madVR when GPU queue is configured lower than 6.
Took me a while to troubleshoot that one as this doesn't happen on NVIDIA (uses fewer reference frames maybe?).Does anyone have a link to 112? Can't find it on the net.Beware that madshi has re-released an updated v112b with the expiration added (for people to compare with results of later builds) so even if you find one on the web, check that it's the original one (files dated from end 2019). It's the same for the official 0.92.17 build btw, it has been re-released to include the new Envy-friendlier licenseSomething in your capture card setup is downscaling the chroma to what looks like quarter resolution (4:2:0)?Yes that definitely looks like something is wrong with the capture. Possibly even both the output device and the capture. Make sure you're not using 4:2:0 as output format, even if HDMI officialy supports it it's shit most of the time.

[fourbigkids]

It's not about the card: you have to downclock it with msi afterburner (limit its top clock speed).
eg my 1060 is limited to 1569 MHz @ 0.800 V, so it won't get hot (~62C) and the fan speed is low (¬750rpm max).
you can use monitor of nvidiainspector to check how your card operates during usage.
about creating complex profiles (https://www.avsforum.com/threads/guide-building-a-4k-htpc-for-madvr.2364113/post-60096583)

Getting a bit out of my comfort zone there chros. :) Will give it a try and report back. Cheers.

edit. As I thought. Too far out of my comfort zone. However, while messing about a bit with madvr settings this morning and re-reading one of huhn's posts about the expense of running more extreme artifact removal amongst other things. I tweaked those and the processing % and temp dropped considerably for the 50i file. Also noticed that they were on for 4K movies as well, disabled the unnecessary 'reduce compression artifacts' and the load on the GPU dropped for dramatically those files. I now have profiles for artifact removal which focus more effort on lower frame rate/resolution video. Cheers.

[flossy_cake]

@fourbigkids glad to see you got it sorted out. Sorry if I mislead you for a bit there -- I was convinced it was the GPU deinterlacing :o

For completeness here is somewhat of a torture test for deinterlacing 1080i. Some of you may have seen this file already as it's quite old.

Clip 1: reference test pattern at 1080p25. (https://drive.google.com/file/d/1rg5VqD3lReXkGFmXxS_8PaIuIZ-KphLs/view?usp=sharing)
No artefacts should be visible here. This is how we want it to look after deinterlacing.

Clip 2: same pattern but at 1080i25 with 2:2 cadence. (https://drive.google.com/file/d/1xdV5_me-pbafnPo6yXhaZKIhe8IP4Ipq/view?usp=sharing)
File has already been tagged with [deint=film] so MadVR should weave fields and retrieve the full 1080p25 and it should look identical to clip 1 if MadVR's film deint is working correctly.

Now tag clip 2 with [deint=video] to see how good your GPU is at deinterlacing it. In MPC-HC you can step through each deinterlaced frame by pausing then ctrl+left/right arrows.

edit: I would love to test the same pattern but with 1080i30 3:2 cadence, as that cadence seems to be problematic with [deint=video]. If anyone has such a pattern please share!

edit2: ok there is something I don't understand. If I play clip 2 with [deint=off] I'm getting the mice teeth / combing artefacts, which is what you get with undeinterlaced 1:1 cadence. But it's 2:2 cadence, so that shouldn't be happening (for instance, see previous 2:2 movie clip from broadcast TV with [deint=off] -- no mice teeth). All I can guess is that maybe the field order was deliberately flipped so as to give a visual clue when deinterlacing is off? Maybe the author of the pattern did that deliberately?

edit3: ah yes I just found madshi confirmed it here 9 years ago (https://www.avsforum.com/threads/hd-1080i-test-pattern-to-determine-vector-adaptive-deinterlacing-others-icl-ticker.1157287/post-22866812) that the field order is deliberately offset by 1 field which is good for testing if the deinterlacer can still handle it correctly.

[flossy_cake]

Random thought: I can't remember what interlaced 1:1 cadence looks like on an interlaced CRT -- does it have mice teeth on moving parts of image? I seem to have a memory that yes it does, but that it wasn't noticeable at a distance?

[mandarynek82]

Where can i get latest test build? It show me that madvr its outdated,

[lvqcl]

https://www.avsforum.com/threads/improving-madvr-hdr-to-sdr-mapping-for-projector.2954506/page-770#post-61878138

[Manni]

Is there a post somehwere with a recap of the issues with madVR and Windows 11 on nVidia (3090)?

I only had a quick look after upgrading and didn't see anything obviously wrong, but then I'm not really using my HTPC to play video content at the moment, so it was a very quick look.

[huhn]

http://bugs.madshi.net/view.php?id=674

you need 20h2 to totally break it.
there are massively move issue i'm not even gonna report.

i can crash madVR by just moving it to another screen by 100%. just as an example.

my win 11 system is AMD.

on my nvidia win10 system i get a ton of crashes where madVR kernel thread or soemthing is supposed to be the issue.

edit: this one should effect you if you are "up to date"
http://bugs.madshi.net/view.php?id=682
multiply user have reported this issue already.

[SirMaster]

I set up madVR (beta 113) for 4k HDR tone-mapping to a JVC RS420 for a friend on Windows 11 and it has been working just find for him for about 6 months so far.

Not sure what problems people are running into.

[SirMaster]

Where can i get latest test build? It show me that madvr its outdated,

Same place as always, madshi's site, increment the number to find new builds as they come out.

https://madshi.net/madVRhdrMeasure163.zip

[huhn]

I set up madVR (beta 113) for 4k HDR tone-mapping to a JVC RS420 for a friend on Windows 11 and it has been working just find for him for about 6 months so far.

Not sure what problems people are running into.
by reading the issue and reproduce it.
saying there is no issue without even testing is just insulting.

[Manni]

http://bugs.madshi.net/view.php?id=674

you need 20h2 to totally break it.
there are massively move issue i'm not even gonna report.

i can crash madVR by just moving it to another screen by 100%. just as an example.

my win 11 system is AMD.

on my nvidia win10 system i get a ton of crashes where madVR kernel thread or soemthing is supposed to be the issue.

edit: this one should effect you if you are "up to date"
http://bugs.madshi.net/view.php?id=682
multiply user have reported this issue already.

Thanks, I don’t use HDR passthrough, I tonemap to SDR. Plus I have nVidia, not AMD. I also only use full screen playback with build 113.

So that will explain why it seems to be working fine here.

[huhn]

on 22h2 the device page will break and starts to map wrong devices.
this means it may even ignore tone mapping.

if you have 2 screen feel free to move it to another when playback is active on AMD the player will black screen forever.

you are on 21h2 so you should also get duplication but in the device tab.

[Manni]

on 22h2 the device page will break and starts to map wrong devices.
this means it may even ignore tone mapping.

if you have 2 screen feel free to move it to another when playback is active on AMD the player will black screen forever.

you are on 21h2 so you should also get duplication but in the device tab.

There was duplication already on Win 10, so that’s not new. I just delete the unneeded devices regularly.

I don’t move the play between two screens. It’s always full screen on the JVC. I’m sure the bugs are real for you, only explaining why I don’t come across the, so why win 11 seems to be working fine here. Plus I’m sure nVidia helps, as this is what madshi is doing all testing on. That and the BT2020 flag support are two of the main reasons why I went for nVidia in my last upgrade.

[fourbigkids]

@fourbigkids glad to see you got it sorted out. Sorry if I mislead you for a bit there -- I was convinced it was the GPU deinterlacing :o

...

All good flossy_cake. Thanks for going to the effort. :)

[fourbigkids]

yes.
that was an excellent advise.


depends all on your what you like too see in the end.
scaler can be very expensive so can error diffusion dithering options.
artifact removal can also be very expensive. i don't know waht you are using and even if i would know it still don't know what you want.




Thanks huhn. Artifact removal settings were indeed expensive and look like the main culprit. I have tweaked them and created separate profiles with much better results, substantial improvement in performance and temp way down. No real change in picture quality that I can detect. Cheers.