madVR - General discussion thread

[mclingo]

you are probably right, i'll pickup something better when I have the cash if its easy enough to do using DISPLAYCAL

[mclingo]

......

Hi, did you get my private message reply, none of them seem to be sending for some reason, my sent box its empty and nothing appears on the message thread.

[Asmodian]

I did. :)

[mclingo]

Thanks again :)

[mclingo2]

Hi, my email account got hacked so I had to shut it down, i'm using this as a templ account, anyone know how I can get my email changed on my main DOOM9 account, there seems to be no support section anymore.

[huhn]

https://forum.doom9.org/forumdisplay.php?f=45

[mclingo2]

i've PM'd foxyshadis

[mclingo2]

I did. :)

Hi, did you get my recent private message form this new account?

Mclingo :)

[MashStari]

Is any of you having issues when playing a movie using PotPlayer+MadVR? After the last W10 updates from yesterday (ver.1903) MadVR crashes when Potplayer starts. :(

[QBhd]

It was fine for me last night... I'm sure I got this week's update

QB

[MashStari]

Well, I installed the last version of MPC-BE and configured with MadVR and works fine, it seems somenting related with Potplayer =/

[Klaus1189]

My advice, stay with MPC-BE ;)

[pirlouy]

I've noticed that with madVR, I have background sound. It's like coil whine, except it is real sound produced by computer.
It means if I use another soundcard that the HDMI (nVidia 1060), for example optical from motherboard, I still have this background sound, and like people mentioned in the past, with NGU, it is amplified.
Sound is really coming for speakers and not motherboard. So strange. Without NGU, it is weak but it is here. I don't have that with mpv ou a browser or even powerful games !

[huhn]

and how is this opotical device powered?

it's impossible to create audio from a video renderer but is is not impossible to fill the signal with dirty power that creates hearable artifacts in speaker.

[Asmodian]

madVR is an unusual power load on GPUs (very spiky) so some GPUs that do not normally whine will when running madVR. Your issue sounds similar. You must have some electrical crosstalk between your audio channels and your GPU. :scared:

[huhn]

this is a pretty far spread issue i can easily make a recording with that using different soundcards one that totally fixed the issue too.

the AVG user doesn't care about stuff liek this they only care about loud...

[pirlouy]

I never said the video renderer created sound by himself. I have no doubt there's no sound code in madVR.
But still I'm surprised how a GPU can create disturbance for something which does not seem that connected (optical soundcard from motherboard).
And I'm also surprised the noise comes only with madVR. I'm certain you will say madVR code is perfect, but as I don't have this sound with other applications (including GPU stress tests), I have my doubts.

the AVG user doesn't care about stuff liek this they only care about loud...I'm not sure to understand this sentence, from my post, we can see I care about sound (and even hardware). I care more about sound that image. I don't use madVR that much, that's why I did not complain before.

[huhn]

I never said the video renderer created sound by himself. I have no doubt there's no sound code in madVR.
But still I'm surprised how a GPU can create disturbance for something which does not seem that connected (optical soundcard from motherboard).
And I'm also surprised the noise comes only with madVR. I'm certain you will say madVR code is perfect, but as I don't have this sound with other applications (including GPU stress tests), I have my doubts.
is the amp DAC power by USB?

GPUs can do what ever they want it it should never effect audio everything else is a hardware flaw.
coilwhine on a GPU is a hardware flaw the program in question has nothing todo with that.
hearable noise in the audio signal is the same.

I'm not sure to understand this sentence, from my post, we can see I care about sound (and even hardware). I care more about sound that image. I don't use madVR that much, that's why I did not complain before.

why would you think i mean you with that only saying this is not as rare as it may seem.

[clsid]

It is called electromagnetic interference (EMI) and it's a pretty common issue. Google for that to get tips on how to reduce/solve it.

[Asmodian]

And I'm also surprised the noise comes only with madVR. I'm certain you will say madVR code is perfect, but as I don't have this sound with other applications (including GPU stress tests), I have my doubts.

As I mentioned before, madVR is a very unusual load on GPUs. A GPU stress test is nothing like madVR. Also, how would 'imperfect' code in madVR cause random noise on your soundcard that changed with the resize algorithm used? And only on your system? It makes a lot more sense that the exact behavior of the GPU power draw, which changes based on the algorithms used and is unique to madVR, is interfering with the audio device.

[pirlouy]

@huhn: No. The DAC is in a dedicated preamplifier (which handles HDMI and optical).
I can understand Coil Whine is a a defect on hardware. But I have my doubts on real sound produced by faulty GPU. The noise is the same with HDMI (from GPU) and optical (from motherboard).

@clsid: Thanks for the exact terms to google. Unfortunately no easy solution in the end.

@Asmodian: I'm still unsure how madVR can be an unusual load on GPU compared to a stress test app. I understand it's easier to think it's an hardware issue for logical reasons. But this sound, really loud when using NGU maximum possible algorithm, is also present when I use lesser algorithms, like Lanczos3. It's not as loud, but still I can hear it if there's no sound at that moment.
And I'm surprised I can hear it with what looks classical algorithms, whereas I don't have this problem with EVR custom for example, or mpv (with directX output).

In the end, instead of searching for solution, I prefer to use mpv with works very well in 95% of my videos.

[huhn]

link me the device plz an optical signal alone can't transfer this. transferring the "dirty signals" sadly happening with HDMI quite often.

as i said i can easily demonstrate this issue and i even have a sound card that can beat it and i have an "expensive" card that worse then on board.

this is not a madVR issues this happens all the time in computer games too is known in application that run at a real high refreshrate like.

case or AVR grounding often helps it can create it too is known as ground loop feedback you can read more here: https://en.wikipedia.org/wiki/Ground_loop_(electricity)

edit: fixed the link doom9 doesn't like links that end with an ")"

[pirlouy]

The dedicated processor is a Nakamichi AV1. It's a rebrand of a Chinese product from Tonewinner company.
The same dirty signal is in my speakers when using optical yet.
If I change source on my processor (and still using madVR NGU in the same time), I don't have this sound on my speakers, so it's not an electrical interference. The sound is really created by some hardware in my computer (GPU/motherboard/PSU).
If it was easy to test with an external soundcard, I'd do it but it's too complicated (limited access + laziness). One day I'll try though.

[huhn]

are you using opt 2 just to make sure you are not using koax? hdmi isn't used at the same time because this alone can create this issue so direct connected to the TV?

If it was easy to test with an external soundcard, I'd do it but it's too complicated (limited access + laziness). One day I'll try though.

don't waste your money i brought a relative bad x570 mainboard because the onboard audio solution was supposed to be far above AVG and it was for me i don't need an soundcard anymore to fix these issues but a friend brought the same board for the same reason and he got is really really bad the issue was present all the time with below 60 FPS games...
the is currently using the audio output of it's monitor it works so i'm not questioning.

my RX 5700 XT is currently reaching a new high and only in one system i can replace the GPU and it is gone or a essence STX but the AE-5 is worse then the onboard but the x570 onboard doesn't care so no conclusion at all i can't even blame the card could be the combination of PSU and card the old board or what ever...

[Asmodian]

If I change source on my processor (and still using madVR NGU in the same time), I don't have this sound on my speakers, so it's not an electrical interference. The sound is really created by some hardware in my computer (GPU/motherboard/PSU).

Why does changing the audio source on your processor make you think it is not EMI? That does not seem to follow to me.

[pirlouy]

I'm not sure what you call "opt 2", but yes, it's optical, not coaxial. When I use optical, TV is off, I use my monitor using DVI.
Tbh, I already have an external soundcard (using USB, not PCIe like in your examples). I'll test it one day but it's really not a priority right now, since I don't have problems with mpv.
Using the DAC from a monitor, wow, even if I would certainly not hear differences, it would not be an option from me. If it's fine for your friend, that's cool !

I often read about Coil Whine (which I don't have), but I never heard of audio interferences too ! That sucks.
Maybe one day Madshi will have this problem and he'll find the "function" which causes this (even if there's no solution in code)...

[huhn]

the biggest problem is the "fix" for your system could create it for other systems. that'S why there is nothing to fix a GPU is simply not supposed to do that but they sadly do...

the next problem madshi is writing mostly shaders these are not close to the hardware and the (plz correct me if i'm wrong btw.) GPU driver has to translate this shader to code the GPU understands which can be different in the next GPU driver even between GPUs while the result is still 100 % the same.

[pirlouy]

@Asmodian: I meant if GPU created an electrical interference (I call it that way but maybe it has nothing to do with EMI, lack of physics skills and vocabulary), even if I changed source, I would hear noise in my speakers.
But yes, it's possible this "EMI" is caused by the GPU and creates dirty signal inside my computer (not outside, that's what I tried to explain). Still, this dirty signal is created only when madVR is in use. Strange.

@huhn: I don't know exactly how shaders are used by GPU and that's one thing I regret about GPU stress tests. They all seem to use same basic functions, but I'm not sure they use shaders like madvr does. That would be cool if there were GPU benchmarks for known shaders (like upscaling algorithms) with several libraries (openGL, directX, Vulkan).

[dwaleke]

@Asmodian: I meant if GPU created an electrical interference (I call it that way but maybe it has nothing to do with EMI, lack of physics skills and vocabulary), even if I changed source, I would hear noise in my speakers.
But yes, it's possible this "EMI" is caused by the GPU and creates dirty signal inside my computer (not outside, that's what I tried to explain). Still, this dirty signal is created only when madVR is in use. Strange.

If the problem you hear is anything like mine then yes it is caused by the GPU when under load. The "coil whine" sound coming through the speakers only exists while the PC is selected as the source on the AVR.

You will not hear this coil whine through the speakers if you have another source selected.

The only way I have found to eliminate the "coil whine" noise from the speakers is to lift the ground on the PC. The coil whine sound still emanates from the video card itself - just not through the speakers any longer.

I should mention lifting the ground is not safe and could cause electrocution.

[Alexkral]

Probably a question for madshi, but there it goes anyway: Does anyone know any detail about how NGU resamples the chroma? I ask it because in some tests I'm doing with Avisynth, just doubling the resolution always produces worse results than doing the color format conversion, no matter the algorithm used. Maybe he has trained some versions of the network specifically for this?

Actually for a 4:2:2 input the OSD doesn't say a word about chroma height downscaling after supposedly doubling the resolution (?)

[huhn]

there is no special NGU chroma scaler only bilateral and reconstruction are specially designed for chroma.

do you remember to fix the chroma position if you are scaling chroma separately in AVIsynth?

Actually for a 4:2:2 input the OSD doesn't say a word about chroma height downscaling after supposedly doubling the resolution (?)
a doubling algorithm usually can do doubling in one way too nnedi3 was actually design to only do that but people found out that if you run nnedi3 again after turning the image 90° turns it into an image scaler.

a chroma NGU version is planned if i remember correctly to replace reconstruct.

the OSD could just lie and do the downscale anyway.

[Alexkral]

This is what I'm doing:

Clip = FFmpegSource2("YV12.mkv")

Y = ExtractY(clip)
U = ExtractU(clip).LanczosResize(clip.Width, clip.Height)
V = ExtractV(clip).LanczosResize(clip.Width, clip.Height)
CombinePlanes(Y, U, V, planes = "YUV", source_planes = "YYY", pixel_type = "YV24")

Except for something I can't see this should be the same as NGU does, however the PSNR is always worse than just using ConvertToYV24.

a doubling algorithm usually can do doubling in one way too nnedi3 was actually design to only do that but people found out that if you run nnedi3 again after turning the image 90° turns it into an image scaler.

I didn't know that about the design of nnedi3, I think it makes sense that NGU is doing the same.

[huhn]

i don't see you fixing the chroma position.

[Alexkral]

Yes, shifting the chroma half pixel to the left improves things, but still worse than Convert, so I don't think it's the main reason.

[Asmodian]

And down? MPEG2 chroma placement looks like:
https://i1222.photobucket.com/albums/dd496/asmodian3/YUV-YCbCr-format-with-2-420-sampling-62.png

[huhn]

moving the chroma by 1/2 pixel to the right is enough after scaling.
pre upscale you have to move is by 1/4.

DSMPTE DV-PAL is top left placed so it needs to be moved by half a subpixel to the bottom and right after scaling 1/4...

for some reason i have top right in my head for mpeg2 but that doesn't seem to be correct.

[Alexkral]

And down? MPEG2 chroma placement looks like:

That's what I thought, but the result is worse shifting up or down. I think Extract correctly arranges the lines, I don't know the reason for the horizontal displacement. The input is a PNG converted with Avisynth to YV12 and back to RGB, and then converted to yuv444p with ffmpeg because for this I'm using VQMT that needs y4m. The PSNR is higher with half pixel shift to the left, same before or after scaling.

[nevcairiel]

A chroma-aware scaler would ideally do the move during the scaling operation, which I imagine madVR does well, since better chroma was one of the main selling points back from the very first madVR version. No experience how to properly reproduce that in AviSynth.

[huhn]

it's always 1/2 a pixel 1/4 is nonsense sorry about that.

[Alexkral]

Well, you weren't completely wrong!! I was shifting 1/2 pixel to the original image or the result. It turns out that the way to do it is to shift the extracted planes when scaling as nevcairiel says, and in this case it is 1/4 pixel because they are half the resolution. :p

This is the resulting script, which gives exactly the same result as ConvertToYV24:

Clip = FFmpegSource2("YV12.mkv")

Y = ExtractY(clip)
U = ExtractU(clip).LanczosResize(clip.Width, clip.Height, 0.25, 0, clip.Width / 2, clip.Height / 2)
V = ExtractV(clip).LanczosResize(clip.Width, clip.Height, 0.25, 0, clip.Width / 2, clip.Height / 2)
CombinePlanes(Y, U, V, planes = "YUV", source_planes = "YYY", pixel_type = "YV24")

[huhn]

1/4 luma is 1/2 chroma so yes could be correct.

[Zenvota]

Is the convergence correction feature going to be added to madvr or is it going to be exclusive to the envy now?

[mark0077]

Does anyone know of a way to make madVR custom display modes, use RGB/YCbCr of different bit depths. I asked this over in the madVR / LG thread but maybe someone here knows. Basically in summary, some modes like YCbCr 444 8-bit limited for me is looking best for 2160p23 SDR, but for 2160p23 HDR I will want to set it to RGB 8-bit Full.

Currently when using custom modes, I cannot see any control over these, so I always get RGB 8-bit.

[shimarin]

Which media players with MadVR can navigate DVD images?

[el Filou]

I've never noticed any issue with madVR and DVD menus.
Here it works perfectly under MPC-HC and MediaPortal, with LAV decoder in DXVA2 copyback mode and madVR in windowed mode.

[el Filou]

Has anyone running Windows Insider builds on NVIDIA Pascal+ or Intel Skylake+ by chance tested the new "hardware-accelerated GPU scheduling" (https://www.minitool.com/news/win10-v2004-hardware-accelerated-gpu-scheduling.html) to see if it gives any benefit with madVR?
What this changes exactly is pretty vague and I don't know what sources those websites that explain what is is are using.

[Audionut]

I finally couldn't help myself and got curious on this update.

Need to join MS insider program to get the preview builds. I chose the slow update channel here (semi-safe I guess).
For Nvidia you need to register on the developer site to access the latest preview driver (450.82): https://developer.nvidia.com/directx
Intel drivers auto updated via MS update to latest WDDM 2.7 drivers.

Works like a charm. No proper testing here but I can defiently push things further with madVR.

[composter]

Ever since I upgraded to Windows 10 all my HDR videos that have the MadMeasureHDR measurements file next to it play almost completely black. Any ideas why?
I upgraded to latest beta (113 I think) and done the HDR measure again and no difference.
If I play the video without measurements file it plays fine with its on-the-fly HDR to SDR conversion, but the picture is obviously not ideal that way.

[SamuriHL]

The measurement files haven't worked in quite a few versions. Disable it in the options for now. The real time is better.

[composter]

The measurement files haven't worked in quite a few versions. Disable it in the options for now. The real time is better.

Thanks for the info.

Is real time really better? Just with default settings for all HDR videos? Is there some kind of tuning/optimization recommended?

Also, is it possible yet to play HLG files yet? Even just in SDR mode?