madVR - high quality video renderer (GPU assisted)

[Sunspark]

I'd call it robin's egg blue, even though it's a lighter shade. Could also argue for turquoise.

[TheGameMaster]

Hi,

No mater what I set doubling too, it shows up as using Super-XBR under control-j. I had it set to ngu-anti. And it just started using super-xbr on its own.

Has this happened to anyone, and how did you fix it? I uninstalled madvr, rebooted, reinstalled, rebooted - same issue - no change.

Rodney

[huhn]

can you share a screen of the osd?

[flossy_cake]

no...
a TV and video only has one 255 0 0 there are no other colors at that position this counts for all primaries.

According to xyY colour model you can keep subtracting 1 from the red channel down to 1,0,0 and all those colours have the same xy coordinates on the CIE chart, only the Y changes.


here is yet another example if that doesn't make it clear i give up.
169 255 255 to 255 255 255.
this is turning this bluish color

Yeah because green and blue channels are clipping in that scenario. In 1000 nits PQ it would be something like 508,768,768 and there is headroom up to 1024 so we'd have enough headroom for a 1.333 multiplier = 677,1024,1024 and according to xyY colour model the xy coordinates should stay the same.

[Amuat]

I came across an SDR video with BT.2020 color primaries for the first time. Is it really so exceedingly rare, or is this just an unlikely coincidence? Anyway, with such videos is it better to turn on HDR manually, keep SDR but change color gamut from standard to wide, or just play it like a BT.709 video?

[Sunspark]

If you've told the renderer what colour primaries/gamut your display uses, it will just re-map.

[flossy_cake]

I came across an SDR video with BT.2020 color primaries for the first time. Is it really so exceedingly rare, or is this just an unlikely coincidence?

What is the title of the video? We might be able to infer something from that.

What matrix is it flagged as? If it's 601 then I reckon the 2020 gamut is probably a mastering error caused by whoever made the file.

I recently watched old short film called Rubberface (1981) that is a NTSC DVD rip but whoever made it incorrectly flagged it as BT.2020 in the metadata resulting in oversaturated colours. It was easy to correct by tagging the filename with [primaries=170M] [matrix=601].

Another possibility is the author of the file intentionally flagged it as 2020 as a way of making the colours look more saturated when played back on a regular 709 display.

edit: you could also make hotkeys to toggle the gamut and matrix while viewing the video to see which one looks visually "correct", that's what I did with that movie to be confident it really was a mastering error. I think you will need to set "this display is already calibrated" in madVR settings in order to unlock the realtime gamut control hotkey (otherwise madVR overlay says "gamut control disabled")

[tp4tissue]

I'd call it robin's egg blue, even though it's a lighter shade. Could also argue for turquoise.

Color of the egg can change depending on the bountifulness of their diet that season.

[Amuat]

I actually use MPCVR nowadays, but since the MPC-HC forum is practically dead I decided to post here, hope you don't mind. The file was a Criterion BD of Ikiru. Here's a link (https://github.com/AmohtAmuat/Ikiru.git) to screenshots of MediaInfo and MPC OSD if they are of any help.

And yeah, obviously it didn't matter in this case as the film is in B&W. That's why I wrote "with such videos", in case I come across a color video of the same type in the future.

[flossy_cake]

I actually use MPCVR nowadays, but since the MPC-HC forum is practically dead I decided to post here, hope you don't mind. The file was a Criterion BD of Ikiru. Here's a link (https://github.com/AmohtAmuat/Ikiru.git) to screenshots of MediaInfo and MPC OSD if they are of any help.

And yeah, obviously it didn't matter in this case as the film is in B&W. That's why I wrote "with such videos", in case I come across a color video of the same type in the future.

Mediainfo suggests you have a legit remux of the 4k UHD Bluray version of that movie since only 4k UHD Bluray supports HEVC, 10bit, Rec2020 matrix & gamut.

I don't believe there is any requirement in the 4K UHD Bluray spec that says only HDR can use 2020 matrix & gamut. As far as I can tell the 2020 spec itself (https://www.itu.int/dms_pubrec/itu-r/rec/bt/R-REC-BT.2020-2-201510-I!!PDF-E.pdf) is actually an SDR spec as it specifies the same SDR gamma encoding as 709. HDR just borrows its gamut, it seems.

Or, possibly the software used to make the remuxed file has incorrectly flagged it as 2020 - does Mediainfo say what application was used to make the file?

[Alexkral]

BT.2020 is for UHD, both SDR and HDR, all Criterion UHD Blu-rays use it as far as I know.

[TheGameMaster]

I coudln't figure out how to paste an image here. But it's showing NGU-Antialias for chroma, as I have it set for that. But for upscale it says super-xbr, but I've got NGU-Antialias setup. I tried uninstalling madvr, clearing the registry, rebooting, then installing madvr - no change. It just started doing this on its own, without me changing anything. No new drivers, nothing. No updates from windows. I havne't tried DDUing the video drivers yet. I'm using two video cards - an RX 5700XT AMD card, and an Nivida RTX 1060. I use the Nivida card to run CUVID copy-back in LAV Video. That's the only thing I can think of that might fix it.

[huhn]

backup your settings.bin and try these: https://drive.google.com/file/d/1Cf0RsvJCOLi6I1dDBJRJ3RW6VGQ0qAgV/view?usp=sharing edit these settings are not meant to be used they are to force NGU AA.
also add showrendersteps as a folder in the settings bin folder that will give you better information of what is happing.

this is not the first time someone brought that up.

but i'm very confused why you would do this:
"I use the Nivida card to run CUVID copy-back in LAV Video."

i mean i just ripped my 5700 xt out of my system and put an old 1060 in because enough is enough but still.

[tp4tissue]

BT.2020 is for UHD, both SDR and HDR, all Criterion UHD Blu-rays use it as far as I know.

I think the reason they use it is because TVs get confused and disrespect the P3 tags.

[TheGameMaster]

Yeah, I have showrendersteps turned on so I can see what deblocking, etc. is doing when I ctl-j. But nothing is out of the ordinary in my setup. Like I said, it just started doing this out of the blue. I have a higher end QLED 65" 4K Sony - my eyes are too old and my room is too bright for OLED (and they don't last a long from what I've gathered - I'm allergic to spending money). I don't like how the TV upscales, so I do all of it on the PC. I run the audio on the Nivida GTX1060ti's HDMI to my Onkyo 9.2 set up, and the video on the RX5700XT. So that leaves the 1060's GPU for hardware decoding using I prefer CUVID copy-back, but have used D3D11 copy-back as well to try to save some frames. I've played around with MadVR since it came out, and prefer MPC-BE over MPC-HC or Pot, MVP, etc. I run LAV for everything, except AAC audio I use ffdshow tryouts (which hasn't been updated since 2014, but it works 1000 time better than LAV for expanding AAC for tricking DTX:X), and run XySubFilter for subs. Still, with all that playing around all these years - I still am a novice when it comes to forcing MadVR outside the gui to do stuff or using scripts for specific video types - I prefer to just set it and forget it at the highest quality with the lowest frame drops. Since its all HDR, I can't use Exclusive mode anymore (thanks windoze - grrr). I did have to set the scale to 1440p for 50fps videos coming from 720p source, instead of full 2160p, to keep it from losing frames. The standard video outside of the video player is 30fps 12-bit RGP Full HDR. I don't have the display options allowed above 1440p50 as it messes with the non-video player video settings. I don't play video games. So, my system can't be THAT high end having to do that. The PC is about 5 years old now, Ryzen 9 3900XT CPU on an MSI Aorus Master 570 board, 64 gigs of RAM. Old school by today's standards. Anyway, I never use the keyboard shortcuts or the system tray, other than ctl-j for osd and ctl-w for subs. I wonder if there is a keyboard shortcut that got switched that could cause it to always load super-xbr instead of what is in the settings? I know there is a stupid clock one that I always have to look up to turn off if the keybord goes flying off my armchair and punches stuff. I wonder if it could be something like that? I tried DDUing the video drivers in safemode, and reinstalling them. And that didn't do anything except waste my time. I tried re-installing MPC-BE, and LAV, and no change. It's really strange that I can't fix it.

[TheGameMaster]

Yes, it was me back on September 29th of last year that had the same issue. You can see the screen shots on that post (2775 or something like that). According to what I said there, it looks like the only fix was an OS upgrade! Yikes!

Rodney

[Amuat]

Or, possibly the software used to make the remuxed file has incorrectly flagged it as 2020 - does Mediainfo say what application was used to make the file?

Sorry, already deleted it due to scarcity of space at the moment.

[whitestar999]

I am getting 3060 12GB & 4060 8GB for the same price so which one should I get only for madvr (no gaming). 4060 is more power efficient but 3060 has more cuda cores.
@huhn @tp4issue @flossy_cake

[ryrynz]

I am getting 3060 12GB & 4060 8GB for the same price so which one should I get only for madvr (no gaming). 4060 is more power efficient but 3060 has more cuda cores.

4060 easy, it's not about Cuda core count but how much faster they are and their relative performance and factoring in the feature set (AV1 decode)

[tp4tissue]

For gaming w/ raytracing 4090 is the only card you should buy, best price to performance ratio if you include the cost of the base PC.

For gaming w/out Raytracing, AMD 7900xtx.

For madvr only, buy AMD, anything after 6xxx which has hdmi 2.1, cheapest card will do, probably 6650xt.

If you believe in Rumors of refresh happening this year. Wait till black friday.

[Klaus1189]

But beware of RX5000 and RX6000 cards if you want to use hardware deinterlacer (needed for correct handling of MBAFF content). Better use Nvidia or RX 7000 card, or even integrated AMD based on RX 7000 (RDNA 3.0)
https://www.techpowerup.com/gpu-specs/?architecture=RDNA+3.0&sort=generation

[tp4tissue]

I can't think of a single good reason to buy nvidia for exclusive madvr use. Overpriced, broken desktop color correction more than half the time, broken dithering more than half the time.

The industry is also in a major sales slump right now, and they're pushing release dates as far as possible, likely into Q1 2025.

Big shakeup in price possibly, as all these 4060/4080 become worthless, well not worthless, probably $500-600 for 4080.

[whitestar999]

For gaming w/ raytracing 4090 is the only card you should buy, best price to performance ratio if you include the cost of the base PC.

For gaming w/out Raytracing, AMD 7900xtx.

For madvr only, buy AMD, anything after 6xxx which has hdmi 2.1, cheapest card will do, probably 6650xt.

If you believe in Rumors of refresh happening this year. Wait till black friday.

I can't think of a single good reason to buy nvidia for exclusive madvr use. Overpriced, broken desktop color correction more than half the time, broken dithering more than half the time.

The industry is also in a major sales slump right now, and they're pushing release dates as far as possible, likely into Q1 2025.

Big shakeup in price possibly, as all these 4060/4080 become worthless, well not worthless, probably $500-600 for 4080.

I am getting 4060 for US $240 & RX7600 for $262. In my region the nvidia cards also have better after sales support than amd cards.

[flossy_cake]

I am getting 3060 12GB & 4060 8GB for the same price so which one should I get only for madvr (no gaming). 4060 is more power efficient but 3060 has more cuda cores.
@huhn @tp4issue @flossy_cake


I would say both are about just as good as eachother for madvr usage.

As far as I can tell the 4060 is only better in modern "triple A" games except for a few edge cases where the 8GB limitation may cause some irregular frame pacing in open world games, but that's a gaming scenario you're not interested in.

The only reason I went with a 3060 was cause I wanted backwards compatibility with windows 7 (4060 has no driver for windows 7) but this became a moot point as there is now just too much stuff incompatible with windows 7, so in hindsight I should have just got the 4060.

But, if I could have bought a 3050 for significantly cheaper I would have gone with that as it has enough power to satisfy my personal requirements for HTPC like up to 4k24.

[flossy_cake]

But beware of RX5000 and RX6000 cards if you want to use hardware deinterlacer (needed for correct handling of MBAFF content). Better use Nvidia or RX 7000 card, or even integrated AMD based on RX 7000 (RDNA 3.0)
https://www.techpowerup.com/gpu-specs/?architecture=RDNA+3.0&sort=generation

I think chroma alignment may be wrong with nvidia gpu deinterlacing. It was wrong on my 1070 but that was with a 4 year old driver on windows 7, maybe 3060 with latest driver will be better? Will test soon, unless someone else can set the record straight.

"automatic source type detection" is working very nicely on my old r9 380 for DVD remuxes tho, it switches nice and seamlessly between ivtc and deint without combing or stutter. Almost makes me consider using it instead of avisynth realtime CPU deint, but I'm too invested in my avs scripts

[huhn]

I would say both are about just as good as eachother for madvr usage.

As far as I can tell the 4060 is only better in modern "triple A" games except for a few edge cases where the 8GB limitation may cause some irregular frame pacing in open world games, but that's a gaming scenario you're not interested in.

The only reason I went with a 3060 was cause I wanted backwards compatibility with windows 7 (4060 has no driver for windows 7) but this became a moot point as there is now just too much stuff incompatible with windows 7, so in hindsight I should have just got the 4060.

But, if I could have bought a 3050 for significantly cheaper I would have gone with that as it has enough power to satisfy my personal requirements for HTPC like up to 4k24.
the 4060 is massively faster easily 15 %. and is still a bad buy... like pretty much all low end GPU priced in the high end.

"automatic source type detection" is working very nicely on my old r9 380 for DVD remuxes tho, it switches nice and seamlessly between ivtc and deint without combing or stutter. Almost makes me consider using it instead of avisynth realtime CPU deint, but I'm too invested in my avs scripts
screenshoot or didn't happen.
this never work at all and the GPU and GPU driver have nothing todo with this.

so if you got this work that would be cool.
I think chroma alignment may be wrong with nvidia gpu deinterlacing. It was wrong on my 1070 but that was with a 4 year old driver on windows 7, maybe 3060 with latest driver will be better? Will test soon, unless someone else can set the record straight.

nvidia hardware processing is still placing the chroma wrong:
d3d11: https://postimg.cc/mtzWnmpX
DXVA2: https://postimg.cc/PppMH8Gj
SHADER: https://postimg.cc/PLFM0QRy

and colors are now messed up great...
edit: it's my clipping shader for madVR 3D LUT ignore it.

[flossy_cake]

still a bad buy... like pretty much all low end GPU priced in the high end.

Performance per dollar is at the top: https://www.techpowerup.com/review/msi-geforce-rtx-4060-gaming-x/33.html

3060 & 4060 are most popular GPU of steam users: https://store.steampowered.com/hwsurvey/videocard/

I agree it's still overpriced, for this price it should come with 10 year warranty, or at least 5 as a bare minimum

the 4060 is massively faster easily 15 %

In "triple A" games at 1080p, yes. At 1440p 12%, at 4k 8%.

Source: https://www.techpowerup.com/review/msi-geforce-rtx-4060-gaming-x/32.html

But madvr is not 3d gaming, I think the difference will be less in a purely 2d shader load but need a 2d benchmark to confirm

But let's say 15% faster in 2d shader, ok that means when playing 4k60 HDR with those settings I mentioned GPU load would be something like 42.5% instead of 50%. With 4k24 HDR , 17% vs 20%.



screenshoot or didn't happen.
this never work at all and the GPU and GPU driver have nothing todo with this.

When using "automatic source type detection" it seems madvr is the code making the decision about whether to output the GPU's deint frame or ivtc frame — is this correct?

I tested a few of my DVD remuxes which have dynamic field cadences where on scene change it switches between 3:2/2:2 & 1:1, and switching between deint and ivtc was seamless. I can frame step across the scene change frame and see it change to deinterlacing mode (otherwise there would be visible combing).

For this to work I have to set LAVs two deint options to "auto", madvr to "auto source detection", madvr to "when in doubt deactivate deint".

In this case for my own 30i and 25i DVD remuxes madvr outputs at double rate 60p and 50p so the downside is you get 3:2 judder at 60p but inspecting the frames one by one it's definitely ivtc by looking at fine details moving across the screen are not aliased it's definitely weaving to 480p and 576p for my PAL DVD remuxes.

A screenshot would only prove chroma alignment, do you have an AMD GPU to test on? I will send you a couple of test clips from my DVDs which I made using MakeMKV, so it's mpeg2 stream inside mkv container with all the correct interlaced metadata for LAV to use

[flossy_cake]

GPU driver have nothing todo with this.


I suppose it depends what IVTC frames the GPU is putting out... they could be progressive frames, but stuttery due to cadence breaks in the source. Maybe AMD driver has better cadence detection for its ivtc mode? Even on animation cadences like 8:7 and 5:5 it's still giving me progressive correctly weaved frames. It wasn't like that on my 1070, NVIDIA driver would output half the frames as bobbed, as if to imply it couldn't detect anything other than 3:2 and 2:2. I will need to do a lot more testing on this with up to date drivers to be sure.

[huhn]

Performance per dollar is at the top: https://www.techpowerup.com/review/msi-geforce-rtx-4060-gaming-x/33.html
performance per dollar from 2023 couldn't more useless.
3060 & 4060 are most popular GPU of steam users: https://store.steampowered.com/hwsurvey/videocard/

I agree it's still overpriced, for this price it should come with 10 year warranty, or at least 5 as a bare minimum
yes the 3050 is so utter trash no one buys it and because there are no cheaper GPUs that's the one people buy.


In "triple A" games at 1080p, yes. At 1440p 12%, at 4k 8%.

Source: https://www.techpowerup.com/review/msi-geforce-rtx-4060-gaming-x/32.html
1080p is fine it will never run out of vram for madVR usage for gaming 8GB can very fast be insufficient for UHD even 1440p

But madvr is not 3d gaming, I think the difference will be less in a purely 2d shader load but need a 2d benchmark to confirm
it kind of is similar to gaming it is literally using the same tool.
game use more stuff of a GPU but shader are an important part of game rendering.
But let's say 15% faster in 2d shader, ok that means when playing 4k60 HDR with those settings I mentioned GPU load would be something like 42.5% instead of 50%. With 4k24 HDR , 17% vs 20%.

if it is 15 % faster it is 15 faster. if you know scale that or not it is just 15%.



When using "automatic source type detection" it seems madvr is the code making the decision about whether to output the GPU's deint frame or ivtc frame — is this correct?
no it doesn't know. my best guess is that it uses ivtc and is not lettign the GPU touch it or just let the GPU touch it.

I tested a few of my DVD remuxes which have dynamic field cadences where on scene change it switches between 3:2/2:2 & 1:1, and switching between deint and ivtc was seamless. I can frame step across the scene change frame and see it change to deinterlacing mode (otherwise there would be visible combing).
every working hardware deint solution will do that.
For this to work I have to set LAVs two deint options to "auto", madvr to "auto source detection", madvr to "when in doubt deactivate deint".
ok.
In this case for my own 30i and 25i DVD remuxes madvr outputs at double rate 60p and 50p so the downside is you get 3:2 judder at 60p but inspecting the frames one by one it's definitely ivtc by looking at fine details moving across the screen are not aliased it's definitely weaving to 480p and 576p for my PAL DVD remuxes.
that mean ivtc isn't used just deint this result is to be expected from a working hardware deint solution.

A screenshot would only prove chroma alignment, do you have an AMD GPU to test on? I will send you a couple of test clips from my DVDs which I made using MakeMKV, so it's mpeg2 stream inside mkv container with all the correct interlaced metadata for LAV to use

yes i have a 5700 XT that does bob bilinear all the time. i don't need to test that anymore.

I suppose it depends what IVTC frames the GPU is putting out... they could be progressive frames, but stuttery due to cadence breaks in the source. Maybe AMD driver has better cadence detection for its ivtc mode? Even on animation cadences like 8:7 and 5:5 it's still giving me progressive correctly weaved frames. It wasn't like that on my 1070, NVIDIA driver would output half the frames as bobbed, as if to imply it couldn't detect anything other than 3:2 and 2:2. I will need to do a lot more testing on this with up to date drivers to be sure.

madVR has no clue what is output it need the frames at the CPU to even look at which is also not happing..

the nvidia driver need 3 consecutive frames to work at all before that it just bobs. so the usual way to see that is taking a frame you want to look at then press 2 times on arrow key left then 4 times on arrow key right on mpc-hc.

[Sunspark]

I'm under the impression that AMD deint is busted on all the newer GPUs and drivers.. I can personally confirm it doesn't work on RDNA2 in Linux, and I've seen complaints about Windows as well.

It doesn't comb, the problem is that it makes these strange looking thick blocks like low-resolution stacks of coloured blocks or something.. hard to explain, but it's not something subtle, it's very in your face.

I would definitely exercise caution around AMD if you're looking for deint.

[flossy_cake]

yes i have a 5700 XT that does bob bilinear all the time.


Oh ok, then I guess it depends on the driver and I just got lucky with my ancient R9 380 running on old "stable" enterprise driver which is doing beautiful seamless adaptive double rate inverse telecine where 1:1 cadence sections get mocomp deint'd to p60 and progressive cadences like 2:2, 3:2, 4:4, 5:5, 8:7 etc are being full frame weaved into progressive frames at p60, output from madvr is always p60 just like my DVD player does and my avisynth script too which does double rate ivtc to p60 to keep it all in a p60 container to preserve all the field cadences (its pretty easy you just do a doubleweave before field matching and voila you got p60 ivtc, avoids the decimation step and preserves field cadences).

It's hard to know what's going on under the hood, I thought DXVA had multiple deinterlacing modes like "video" and "film" and those were implemented on the driver side and madvr is choosing between the two when set to "auto source type detection" , but it sounds like my understanding was wrong there and what it really does is something else ... so when you enable automatic source type detection (i.e UNticking "disable automatic source type detection) what is it automatically switching between? Is this determined only once at start of playback based on directshow metadata, or analysed on a per-frame basis?

[whitestar999]

the 4060 is massively faster easily 15 %. and is still a bad buy... like pretty much all low end GPU priced in the high end.


In my country 4060Ti cost 70% more than 4060 so very hard to justify for only madvr usage. 3060Ti here is only available used or very old stock with doubtful warranty period.

Performance per dollar is at the top: https://www.techpowerup.com/review/msi-geforce-rtx-4060-gaming-x/33.html

3060 & 4060 are most popular GPU of steam users: https://store.steampowered.com/hwsurvey/videocard/

I agree it's still overpriced, for this price it should come with 10 year warranty, or at least 5 as a bare minimum


I am getting 5 years warranty on 4060 as well as 3060.

performance per dollar from 2023 couldn't more useless.

yes the 3050 is so utter trash no one buys it and because there are no cheaper GPUs that's the one people buy.



1080p is fine it will never run out of vram for madVR usage for gaming 8GB can very fast be insufficient for UHD even 1440p


it kind of is similar to gaming it is literally using the same tool.
game use more stuff of a GPU but shader are an important part of game rendering.


if it is 15 % faster it is 15 faster. if you know scale that or not it is just 15%.


I got 4060 for $236 with 5 years warranty while friend got 3060 12GB for $226 on my behalf both with 5 years warranty. Friend is not interested in keeping the card & I have decided on 4060 so far so will need to sell 3060.

[flossy_cake]

I got 4060 for $236 with 5 years warranty

That's a good deal, I paid 275 USD (400 AUD) for a MSI Ventus 2x 3060 which only has 3 years warranty by the looks of it. Their stupid website doesn't actually say how many years warranty, have to rely on reddit posts. And out of the box one of the thermal pads wasn't making contact with a VRAM module so that module probably would have burned out after a couple of years if I didn't randomly notice it.

edit: oh and the fan shroud is unusually tall on this Ventus model and presses against the chipset heatsink so it barely fits, and the motherboard is a brand new MSI so I guess they don't check for compatibility between their own products

https://i.ibb.co/SVSrMVf/ca0b425c19ba2dee484d47e48684d6a0.jpg

(it's actually touching, I can't slide a piece of paper in between them)

[flossy_cake]

broken dithering more than half the time.

Yeah I'm just starting to notice this today, coming from Windows 7 the dithering setting was rock solid being applied with Calibration Tools. On Windows 10 I've had a couple of random dropouts.

I've put a shortcut to a grey ramp pattern on the desktop so I can easily check it every few hours and after reboots.

There's some trick to getting it to stick on Windows 10 that I haven't fully figured out. When I was working on Calibration Tools a couple of years ago testing it on Windows 10 I made it so the app disables dithering on exit, so when you shutdown the PC it's in a disabled state before the next reboot, which seemed to be working.

But that isn't working all the time on 22H2, now I've added a startup delay of 20 seconds in the Config.ini to try and give more time for the NVidia driver to initialise on startup before applying dithering and cal (it enables dithering first, then loads the cal... I'll have to experiment swapping this order around to see if that affects it).

Signing in/out affects it too, had a dropout this evening on a cold boot with a 10 second startup delay and signing out/in brought it back.

Putting PC into sleep mode also affects it, I would avoid using sleep, and I don't use fast startup either as that is basically the same.

With a 20 second startup delay in Config.ini no dropouts yet, but I'm not confident.

It also has something to do with whether I've modified the calibration state since the last reboot. It seems like once I do a boot -> wait for driver to init -> enable dithering and cal -> make no modifications to cal state -> restart, from then onwards it seems to be ok?

This is on 22H2, so if you're on an older version this might not apply.

I think Windows Color Management system is interacting with it somehow too, but I'm mostly thinking it's a NVidia driver problem. It used to be that if I opened Video Playback in Windows System Settings (the preview video of a guy holding a skateboard) that would instantly kill dithering, but this is not the case on 22H2 and latest NVidia driver. So something changed.

There may be a race condition going on a shutdown where Windows is shutting down simultaneously to CalibrationTools.exe and maybe it's not getting a foot in the door soon enough to ensure dithering has been disabled long enough before everything else begins to shutdown. Can try manually disabling it by right clicking CalTools system tray and exiting which will turn off dithering (since exiting the app disables dithering) then wait a couple seconds, then shut down.

For me it's holding up like 95% of the time since I increased startup delay to 10 seconds, but I don't want to live with having to check grey ramps every day.

[huhn]

you can make madVR run a preset of colorcontrol and the problem is solved.

[flossy_cake]

you can make madVR run a preset of colorcontrol and the problem is solved.

I am curious to know how that might solve the issue, as far as I know both apps use the Nvidia API to set dithering in the driver so wouldnt both be vulnerable to dithering dropouts due to bug in NVIDIA driver? Or does colour control have some special trick that makes it stick?

In the past I had success by disabling windows colour management in the windows task scheduler but that doesn't seem to have any effect on 22H2.

For me the thing which seems to be having the most effect on 22H2 is having a long startup delay before enabling dithering on boot, but I still had 1 dropout even with a 10 second delay. But I was getting more dropouts when I had no delay at all so it seems related somehow.

Maybe it also needs a longer delay between setdither and setcal because currently I am going win startup -> wait 20 sec -> setdither -> no delay -> setcal. Maybe the driver is still doing it's setdither init routine and I'm loading a cal too fast and it's getting confused or something? I'll try putting a delay in between and see if that improves it.

One thing I have not experienced so far is a dropout caused by an app. So once I am booted to desktop and dithering and cal are loaded, from them on dithering has always stuck at least until the next reboot. Whereas in older windows versions there were certain apps that could break it like even opening the windows appstore would break it, I believe that was on 1903.

So for me the issue seems related to startup and shutdown (and sign in and sign out, and sleep and wakeup) which I suppose is when the NVIDIA driver is doing it's initialisation.

There is another stupid bug with the current NVIDIA driver where if I touch any of the desktop colour controls in the NVIDIA control panel even once, from that point onwards the driver does this thing at startup where it blinks the screen black for about 1 second when first reaching the desktop as if it's doing some extra initialisation that requires reintialising the display mode. Really ticks me off, had to do a "clean install" of the driver to prevent that behaviour. I don't want my display blinking black on every boot, that is really annoying to watch that happen every day when starting the HTPC.

[Sunspark]

What is with Nvidia and AMD that they have all these resources and funds and their drivers are so bad?

[flossy_cake]

What is with Nvidia and AMD that they have all these resources and funds and their drivers are so bad?

It's the competence crisis. Back in the day they had 33mhz CPUs and had to be so intelligent with everything.

But in their defense when a codebase becomes so large and bloated over decades with so many modules being added it can be almost impossible to know what is happening at runtime. It's more like, the bugs are gonna happen, so let's just try to catch as many as possible after the fact.

NVIDIA dithering API isn't publicly documented / officially supported and if we complained about bugged behaviour they would probably just nuke it at the next update. Linux drivers have it tho in the NVIDIA control panel, last time I checked.

When dithering bug happens in windows, dithering is still actually active because if I set it to 6-bit static I see the dither pattern appear/disappear when toggling it. So the problem isn't that it isn't getting enabled, rather the dithering pattern is not aligning with the tones being rendered. I think this is somehow related to Windows colour management system. The dithering needs to be aligned +/- 1 tone around the high precision float otherwise it won't work.

[mclingo]

Has anyone else tried and fluid motion frames the Blue Sky Plugin and media player classic yet?, I was just having applied with it I quite like the results it's kind of like Smooth motion without the artefacts and smearing, Going to watch your whole film with it first before I switch to it but it does seem to play nice with mad VR as well even if it does report that it's going to skip a frame every three seconds but it doesn't.

[ryrynz]

Damn, if it's any good I might actually consider an AMD card.

[mclingo]

I definitely prefer it to smooth motion but I'm undecided whether I prefer the movies playing at the correct frame rates, However I only tried this because my setup doesn't seem to be able to do 24P correctly, I get frame Skips every 40 seconds no matter what I do, all of a frame rates are hours and days I don't know whether it's my pc, amd card on my tv that can't do 24P, I can't work it out

[flossy_cake]

I definitely prefer it to smooth motion but I'm undecided whether I prefer the movies playing at the correct frame rates, However I only tried this because my setup doesn't seem to be able to do 24P correctly, I get frame Skips every 40 seconds no matter what I do, all of a frame rates are hours and days I don't know whether it's my pc, amd card on my tv that can't do 24P, I can't work it out

Skip every 41 sec is the diff between 23.976 and 24.000 , I am 100% sure your LG OLED supports at least the former, so in madvr auto resolution string put 2 modes eg 2160p23, 2160p24 and you will need to have both modes available in windows display settings for it to work otherwise you'll have to use CRU to manually create the missing mode.

In my madvr autores string I have 1080p23, 1080p24 and it's working with test files and ctrl-j will tell you which one it chose. Only a small number of films use 24.0 so I wouldn't stress too much about not having a 24.0 mode so you could just put 2160p23 on its own.

Other than that all I can think of maybe some windows compositor issue?

A longshot is that maybe your 2016 LG OLED has a broken 24hz mode where it does 24.0 instead of 23.976 and possibly CRU might be able to resolve that if that is the issue.

Edit: well the fact you're saying ctrl-j reports 1 frame drop every hours or days and you're still getting 1 visible drop every 41sec might seem to imply you are actually sending 23.976 to the display in which case the problem might be inside the TV where its internally refreshing the screen at 24.0 instead of 23.976 when receiving a 23.976 signal over HDMI. Have you disabled all motion interp settings on the LG?

Another option is to remove the p23 p24 modes from the madvr autores string and just output it as p59.94 which 23.976 is evenly divisible into, but then you will have 3:2 judder.

Edit: according to rtings the LG can convert 60p input signal to 24p at the display so you could try that TV setting as well and see if that makes it come out as 24.0 or 23.976 at the display.

And just to make sure the files you're testing with are definitely 23.976 not 24.0 and you don't have reclock installed?

[huhn]

Has anyone else tried and fluid motion frames the Blue Sky Plugin and media player classic yet?, I was just having applied with it I quite like the results it's kind of like Smooth motion without the artefacts and smearing, Going to watch your whole film with it first before I switch to it but it does seem to play nice with mad VR as well even if it does report that it's going to skip a frame every three seconds but it doesn't.

it just just takes a 23p stream and outputs a 59p stream. throwing half the good frames away. it literally does the same stuff a TV does at motion handling high just worse.

it has nothing todo with smooth motion which is just blending frames and nothing else.

[Alexkral]

I think a very simple solution for some of these problems would be to use AviSynth Filter (https://github.com/CrendKing/avisynth_filter) with a simple script that uses FPS (https://avisynthplus.readthedocs.io/en/latest/avisynthdoc/corefilters/fps.html).

[huhn]

no that does not change that the audio and video clock run at a different speed and don't sync. changing the video speed will not change the speed of the gpus output

reclock fixes it by actually changing the speed.
so would changefps but not assumefps but you can just use smoothmotion instead because it does the same.

[flossy_cake]

I use avisynth for slowing down 25 to 24 in realtime for my pal dvd remuxes, the function for that is Assumefps(numerator=24000, denominator=1001, sync_audio=true) which slows both the video
and audio down resulting in 4% lower pitch audio.

Actually I set sync_audio=false and manage the audio separately with the built in Timestretch() function as it's more flexible eg. slowing down audio without affecting pitch for pal DVDs which had already been pitch corrected in the source eg. Due South PAL DVDs.

And for other ones like Lois and Clark PAL DVD has 50fps special effects so I want to not decimate those so I just pitch correct the audio without affecting its speed so it plays back at 50p 4% too fast but with correct pitch. Cause Timestretch lets you adjust pitch and tempo independently. Avisynth is amazing I'd recommend learning it , I even use it on 1080p stuff for audio peak normalisation or doing downscaling to SD for shows that weren't really designed to be shown in HD in terms of production values

Someone should make a helper GUI app that generates avisynth scripts for you by clicking what features you want and then create a mirror of source folder structure but with .avs files instead of .mkv or whatever. I got this with autohotkey script but it's not user friendly enough, perhaps I will have a go at GUI

Oh and you can do frame interpolation with Interframe() function which is just a wrapper for SVP. It works great on slow pans and stuff but has too many artefacts on complex patterns like actors waving hand gestures and stuff. I only used it on pilot episode of black adder to get it to 50fps to match the rest of the series.

[huhn]

if you are doing sync audio it is do something similar to reclock.
but it is not dynamic.

[clsid]

MPC-HC has that functionality built-in. No need for ReClock or Avisynth.

Menu > play > playback rate > 24 or 25 fps

This adjust playback speed to play 25.0 as 23.976 or vice versa (PAL <-> NTSC). If you use SaneAR or MPCAR as audio renderer, pitch is corrected automatically.

[huhn]

this is under the context of fixing repeated and dropped frames thanks to audio video clock sync issues.

[ryrynz]

it just just takes a 23p stream and outputs a 59p stream. throwing half the good frames away. it literally does the same stuff a TV does at motion handling high just worse.



it has nothing todo with smooth motion which is just blending frames and nothing else.Lame. TV's with ULMB2 when.