View Full Version : madVR - high quality video renderer (GPU assisted)
dejarik
1st October 2020, 13:33
Gtx 1070
huhn
1st October 2020, 13:43
try driver 456.38. state the model of the Tv too.
dejarik
1st October 2020, 13:49
Driver I'm using are the 452.06
Tv Model is a Sony XBR55X900H
I'm gona test the 456.38 and let you know
huhn
1st October 2020, 14:05
HDR generally shouldn't work with that driver.
and i have issues with 456.55 but i had a lot of issues so no need to talk about it.
and you are in luck i know a lot of the xh90
dejarik
1st October 2020, 14:51
Seems to have made a difference i dont get the weird artifacts
but HDR doestn seem to be enable correctly picture is washed out
huhn
1st October 2020, 14:52
try fullscreen exclusive.
dejarik
1st October 2020, 15:29
Did a few test for some raisons tone map hdr using pixel shaders
seems to give better results i have notice that dynamic metadata
is not supported on this Tv don't know if it makes a difference ? i
would think maybe it does
huhn
1st October 2020, 15:46
the TV generally doesn't really care about meta data and madVR doesn't support dynamic metadata AFAIK. it just measures it by a frame by frame basic anyway.
this TV doesn't support dolby vison (at all) properly and can't handle it correctly. not to be confused with the dolby vision on UHD BD which is still HDR10.
dejarik
1st October 2020, 16:02
Tone map hdr seems to gives a verry decent results, also
i should update the firmware on this Tv i'm waiting for the
hdmi 2.1 and 120 Hz software update, should be release about the same time the ps5 comes out.
thanks for your help huhn
senzaparole
1st October 2020, 17:58
@VBB
@clsid
thanks for your valuable advice
DMU
1st October 2020, 21:24
Recently I noticed in the OS HDR mode the metadata became correct.
Pic (https://drive.google.com/file/d/16lAiIuyXnsS338Y7PSftadfIyB_Ffhen/view?usp=sharing)
tp4tissue
2nd October 2020, 01:45
Yeah, FRC is a type a dithering, but I would reckon madVR's error-diffusion dithering is higher quality than your display's FRC algorithm.
I thought madvr's error diffusion dithering remains on even if you set pipeline to 10 bit.
On all the Roku platform TVs I've tested, on Test patterns, 10bit @ 23.976hz always produced smoother gradient than 8bit modes. <setting is 12bit mode on NVpanel, 10bit dx11 surface in madvr>
My samsung Tv won't support 10bit 444 from PC. no go here.
I've got an older 10bit 24" IPS u2410 which also has smoother gradient when set to 10bit mode.
el Filou
2nd October 2020, 12:08
Recently I noticed in the OS HDR mode the metadata became correct.
Pic (https://drive.google.com/file/d/16lAiIuyXnsS338Y7PSftadfIyB_Ffhen/view?usp=sharing)Can you post this in the driver info thread: https://forum.doom9.org/showthread.php?t=176013&page=98 ?Did a fair bit of research when i set this all up so i'm sure it was for a good reason but i can't tell you what that was for the life of me. Lowered it down to 12. W7, yes, no aero and desktop compositing off, but tried turning all of that on and setting it up from madVR and it still defaults to D3D9.Sorry, I've tried everything, switching to Exclusive, D3D9 presentation, and increasing my present queue up to 16, but I can't reproduce this. Exclusive mode works quite differently in Win7 though, even with fullscreen optimisations disabled in Win10, and I don't have Win7 system to test on.
Did you try fiddling with the GPU settings? Exclusive mode is like games, so some GPU settings may influence it. Some stuff like double/triple buffering, enhanced/fast sync, that kind of stuff? I don't game much so don't know about all of them. Maybe do a system restore checkpoint to go back to and then try resetting the graphics driver settings?
Asmodian
2nd October 2020, 15:31
I thought madvr's error diffusion dithering remains on even if you set pipeline to 10 bit.
Of course dithering stays on at 10 bit, but if you then dither again to 8 bit you have double dithering (higher noise) and the quality of the dithering is almost entirely due to the second step.
On all the Roku platform TVs I've tested, on Test patterns, 10bit @ 23.976hz always produced smoother gradient than 8bit modes. <setting is 12bit mode on NVpanel, 10bit dx11 surface in madvr>
My samsung Tv won't support 10bit 444 from PC. no go here.
I've got an older 10bit 24" IPS u2410 which also has smoother gradient when set to 10bit mode.
Are you saying madVR does not have smooth gradients in 8 bit? What about with ordered dithering (the smoothest option)? Your display can convert from 10 bit to 8 bit internally with dithering that is smoother than madVR's? That does not sound right to me, you should have smooth gradients from madVR all the way down to 4 bit. Gradients normally get nosier as bit depth decreases, not less smooth.
Make sure linear light dithering is enabled.
cosmitz
2nd October 2020, 16:20
Did you try fiddling with the GPU settings? Exclusive mode is like games, so some GPU settings may influence it. Some stuff like double/triple buffering, enhanced/fast sync, that kind of stuff? I don't game much so don't know about all of them. Maybe do a system restore checkpoint to go back to and then try resetting the graphics driver settings?
There shouldn't be any conflicts, no triple buffering, vsync app-based, no adaptive sync, maximum performance, and the issue has been there for relatively forever, not even sure i can trace it to anything changed in particular.
tp4tissue
3rd October 2020, 00:45
Are you saying madVR does not have smooth gradients in 8 bit? What about with ordered dithering (the smoothest option)? Your display can convert from 10 bit to 8 bit internally with dithering that is smoother than madVR's? That does not sound right to me, you should have smooth gradients from madVR all the way down to 4 bit. Gradients normally get nosier as bit depth decreases, not less smooth.
Make sure linear light dithering is enabled.
I tested several different conditions with dithering on and off. Using a 10bit gradient test video.
Mode 1
Madvr Dithering set off. All 8 bit pipe set (Nvidia+Madvr), the test pattern 10bit gradient portion across 3840 pixels is choppy 8bit stripes.
M2
Madvr Dithering set off. 12bit NV, 10bit Madvr, 4K@23.976hz the test pattern looked smooth. Slight segmentation visible
M3
Dithering on (erDiff). 8 bit pipe set (Nvidia+Madvr), the gradient IS SMOOTH, however. near the dark tones, I can still see slight segmentation, but the edges are softer. Vs M2, It's smoother than M2, so Madvr's dithering is indeed quite good.
M4
Dithering on (erDiff). 12bit NV, 10bit Madvr, 4K@23.976hz, I can just barely make out the segmentation.
In M4, shadow detail, near black transitions are much better.
I noticed the film grain also looked more pronounced, I can't explain this as I would've thought the dithering would kill that off.
mytbyte
3rd October 2020, 18:19
I can see it's all about bits around here. So I'll jump right in.
I got myself a Lenovo d32q-20 monitor with 10 bit panel (8bit+frc). However, I'm not offered 10 bit color depth from the graphics card, neither RX 580's Radeon Software nor GTX960 Nvidia Panel. However, again, when I play 10 bit gradient test (both mp4 or TIF) through MadVR, I get very smooth gradation in D3D11 full screen on RX 580 (both windowed and exclusive) with MadVR's dithering off, while I do see striping when playback is not fullscreen. Does that mean that 10 bit pipeline is somehow passed through to the monitor avoiding the windows bit depth settings or there is some hardware dithering down to 8 bit by the GPU. I vaguely remember there was some talk about AMD doing the dithering by itself. I'm yet to test the GTX960 but some thoughts on the matter be nice in the meantime.
By the way, did Madshi really abandon MadVR? What's he doing now, some hardware implementation of it?
tp4tissue
3rd October 2020, 21:02
In non-fullscreen mode, it goes to the 8 bit pipe. for example when you pop the menu up, or when you right click.
Hit Ctrl J, and you can see it go in and out on the top left indicator. This has nothing to do with AMD.
AMD's advantage is it does dithering when 10bit is enabled, which means it will be a bit smoother than nvidia. This is especially true if you have a Color correction gamma table loaded. AMD will also do dithering in 8-bit mode, if you have a table loaded.
But this is not useful for Madvr, because if you have a colorimeter, madvr can take care of the entire color pipe more reliably through its 3DLUT color correction support.
mytbyte
3rd October 2020, 22:37
@tp4tissue: I must admit I'm not sure I can follow you about the Gamma Correction color table but I do have a 3D LUT loaded that I had created with DisplayCAL. Gamma table = 1D LUT?
I have just introduced "DP_DisableDither=1" value to the Windows registry's AMD driver section and I now have banding/striping in fullscreen mode as well...so it was not a full 10-bit pipeline file-to-monitor after all but AMD's default dithering. Why the hell this monitor has 10 bit panel when it can't take a 10 bit signal?
tp4tissue
4th October 2020, 00:11
/headscratch
What colorimeter are you using for madvr's 3dlut.
So nothing's changed from before EXCEPT the DP_disabledither ? What are the madvr settings you're testing with.
mytbyte
4th October 2020, 00:34
/headscratch
What colorimeter are you using for madvr's 3dlut.
So nothing's changed from before EXCEPT the DP_disabledither ? What are the madvr settings you're testing with.
X-rite Colormunki.
What specific MadVR settings would you like to know? The monitor properties is set to "10 bit or higher", "enable automatic fullscreen exclusive mode" (though it's equally smooth with windowed fullscreen), D3D11 for presentation, no dithering. Seems like AMD dithering activates in full screen only.
tp4tissue
5th October 2020, 20:24
X-rite Colormunki.
What specific MadVR settings would you like to know? The monitor properties is set to "10 bit or higher", "enable automatic fullscreen exclusive mode" (though it's equally smooth with windowed fullscreen), D3D11 for presentation, no dithering. Seems like AMD dithering activates in full screen only.
It may be the case that madvr's 10bit pipe requires that amd dithering bit to exist, or amd's own pipe requires it..
We need another app which supports 10 bit pipe to test this. I don't think amd works with 10bit opengl on photoshop.
hrrrm...
mytbyte
5th October 2020, 21:46
It may be the case that madvr's 10bit pipe requires that amd dithering bit to exist, or amd's own pipe requires it..
We need another app which supports 10 bit pipe to test this. I don't think amd works with 10bit opengl on photoshop.
hrrrm...
I think I have tested quite extensively now to determine with certainty that the smooth gradient is the result of AMD's own dithering from 10 to 8 bit and not some "phantom" 10 bit pipeline to the monitor that defies the laws of physics. The banding observed with GTX 960 in fullscreen mode further proves that the monitor's input only receives 8 bit color depth and Nvidia doesn't dither 10 -bit to 8-bit.
You are right - even if the monitor were truly 10-bit input-to-output, I wouldn't be able to achieve 10-bit with AMD consumer GPU. And the GTX 960 (Maxwell, prior to Pascal) is exempt from new Nvidia's Studio driver's ability to make use of 10 bit OpenGL used in most productivity apps.
VBB
5th October 2020, 22:30
Just to add to this: Nvidia drivers do not dither when using RGB full. They dither with RGB limited and YCbCr.
mytbyte
6th October 2020, 09:50
Just to add to this: Nvidia drivers do not dither when using RGB full. They dither with RGB limited and YCbCr.
You are very right, sir. Just confirmed this.
And now another thing: I just updated the GTX960 machine to Windows 2004 build and 10-bit and 12-bit options magically appeared, but for HDMI YCbCr 4:2:2 only (DP still 8 bit only). I can't absolutely swear it wasn't there before but I believe I had tried everything and don't believe I missed it.
AMD machine didn't get the 2004 update and is still oblivious to 10 bit in any form.
QBhd
6th October 2020, 12:30
Just a heads up and an FYI... but CRU will also limit output to 8-bit only.
Edit:
Actually I went back and read my post about custom resolutions (https://forum.doom9.org/showthread.php?p=1895131#post1895131) and I guess it would be the lack of HDMI standard timings that could be limiting output to 8-bit...
So I will change my above statement slightly... The EDID can have a big role to play in the bitdepth of the output.
EDIT 2:
@mytbyte, maybe you can go backwards from what I was attempting back then and add those standard timings back in, if they are missing from your EDID...
QB
SirSwede
6th October 2020, 14:29
I am looking to get a Lenovo Yoga S740 15, with a GeForce GTX 1650 with Max-Q graphics card.
I was wondering what this card is capable of, using madVR and MPC-HC, scaling 480p and 720p to it's own 1080p screen?
Will I get any decent quality from this material or shouldn't I be buying a machine with this card?
Thank you in advance! :)
clsid
6th October 2020, 15:33
Yes, that card will work fine.
mytbyte
6th October 2020, 16:22
Just a heads up and an FYI... but CRU will also limit output to 8-bit only.
Edit:
Actually I went back and read my post about custom resolutions (https://forum.doom9.org/showthread.php?p=1895131#post1895131) and I guess it would be the lack of HDMI standard timings that could be limiting output to 8-bit...
So I will change my above statement slightly... The EDID can have a big role to play in the bitdepth of the output.
EDIT 2:
@mytbyte, maybe you can go backwards from what I was attempting back then and add those standard timings back in, if they are missing from your EDID...
QB
Bless you and this forum. In CRU, I just added 30-bit and 36-bit under HDMI support in the Extensions block. Works for both HDMI and DP.
What would be the freaking reason why the monitor doesn't advertise 10 and 12 bit support in EDID?
chros
6th October 2020, 19:01
Maybe it's only 8bit? Can you link where you see it's 10bit? And some panel Ids as well e.g. from madvr.
KrazyLurt
6th October 2020, 21:46
Have anybody more than I experience trouble adjusting 23.976 hz with CRU when using nvidia driver 451.61 and above?
Setting my near perfect values in CRU gives me a problem when setting that custom resolution in NV Cp and hit apply it will revert to the standard 23 hz option and not the 24hz custom option made by CRU.
The only way around that issue is to set a CRU value far enough from the standard 23.976 nv option , but of course the would mean ever higher frame drops or repeats.
I use Philips 55POS OLED TV, a 10bit RGB limited Tv.
mytbyte
6th October 2020, 21:58
Maybe it's only 8bit? Can you link where you see it's 10bit? And some panel Ids as well e.g. from madvr.
Leaflet:
https://psref.lenovo.com/syspool/Sys/PDF/datasheet/Lenovo%20D32q-20_datasheet_EN.pdf
1.07 billion colors mentioned at least twice. ;)
DisplaySpecifications website:
https://www.displayspecifications.com/en/model/5f971f18
Screenshots made after applying CRU:
https://drive.google.com/file/d/1jzY6KAhFF-0-Es3_QqE1xTsQ7Lpj8CIc/view?usp=sharing,%20
https://drive.google.com/file/d/1oKyrvwj2BiCi4RZZQbM9mDFWLg6A5ZKf/view?usp=sharing
Also, with DP_DisableDither=1 and MadVR Dither = None gradients are now visibly as pristinely smooth as I'll ever be able to notice.
chros
6th October 2020, 22:07
Have anybody more than I experience trouble adjusting 23.976 hz with CRU when using nvidia driver 451.61 and above?
Setting my near perfect values in CRU gives me a problem when setting that custom resolution in NV Cp and hit apply it will revert to the standard 23 hz option and not the 24hz custom option made...
Don't use NV cp to change refresh rate, it works like that! Use windows OS one or a util, like Display Changer II (there is a link in the driver thread).
KrazyLurt
6th October 2020, 22:40
Don't use NV cp to change refresh rate, it works like that! Use windows OS one or a util, like Display Changer II (there is a link in the driver thread).
Thanks Chros for the quick reply, I’ve been reading this thread daily for several years but almost never post.
I will use Display Changer II for sure , thanks for the tip , and I agree nv cp is terrible and windows OS doesn’t have color depth.
I guess Display Changer have options for toggle RGB limited mode. If I knew that Philips 55POS didn’t use RGB Full I would have gone with a LG variant.
Anyway hopefully this will address my issue and the CRU custom resolution is useable. I do a fair bit of switching from 60hz to 23.x so a good software like Display Changer II would be awesome.
Last resort is to leave the 23.976 along and use 32-bit Pot Player or Mpc-HC with ReClock.
QBhd
7th October 2020, 04:07
...
Last resort is to leave the 23.976 along and use 32-bit Pot Player or Mpc-HC with ReClock.
Or you can do what I do now and try the "Built-in WASAPI Audio Renderer" in PotPlayer. I used to try and chase the perfect custom resolution or as close to perfect as possible. But what I found is that just using the default 23.976 resolution and the WASAPI renderer the gives either an infinite time between drops/repeats or an interval that is in the hours/days range.
QB
KrazyLurt
7th October 2020, 05:45
Or you can do what I do now and try the "Built-in WASAPI Audio Renderer" in PotPlayer. I used to try and chase the perfect custom resolution or as close to perfect as possible. But what I found is that just using the default 23.976 resolution and the WASAPI renderer the gives either an infinite time between drops/repeats or an interval that is in the hours/days range.
QB
That’s strange because I gave the built-in WASAPI Audio Renderer a new shot yesterday, and while in gave “no expected drops” and a 0.00000 clock deviation in MadVr ctrl+j stats , I still got a dropped frame pretty early on the the MadVR stats.
But I will certainly test it again , because it would be awesome if PP’s built-in WASAPI means frame and audio is at perfect sync ! Thanks again for the heads-up!!!
KrazyLurt
7th October 2020, 15:12
Or you can do what I do now and try the "Built-in WASAPI Audio Renderer" in PotPlayer. I used to try and chase the perfect custom resolution or as close to perfect as possible. But what I found is that just using the default 23.976 resolution and the WASAPI renderer the gives either an infinite time between drops/repeats or an interval that is in the hours/days range.
QB
I re tested this and like above , no deviation or drops/expected but when running the test , madVr still counts drops/repeated frames, I stopped the test after 2 drops.
I think it’s just that MadVR can’t estimate or do accurate measurements when using WASAPI, and that explains my results. I would be great if you do a test as well.
QBhd
7th October 2020, 17:07
Did you see the drops? Do you see then during playback of an entire show/movie? I am not talking about looking at stats, I am talking about actually noticing them happen... Because as I said above, chasing the perfect refresh rate is a lot of effort for something that may not even be noticed. I quit the chase and have been happier ever since :)
QB
SamuriHL
7th October 2020, 17:10
Something else I don't see mentioned (maybe it has and I missed it) but it takes madvr a few seconds to "settle down". I start playing a video and after 2 or 3 seconds reset the stats and so far have noticed no drops after that.
KrazyLurt
7th October 2020, 17:13
No I let run without watching so I totally see your point, I am just curious. :thanks:
rack04
8th October 2020, 01:03
Is there a significant difference between the processing speeds using madVR between Windows 7 and Windows 10? The reason I ask is because I recently "upgraded" to Windows 10 from Windows 7 and I copied over the settings file and the movies I used to watch on Windows 7 with 0 dropped frames now really tax the GPU and result in dropped frames. LAVFilters was installed in both cases.
VBB
8th October 2020, 01:12
Check power profile in Nvidia control panel. Change it from "optimal" to "adaptive". Also make sure LAV decoder is set to "D3D11" and "Auto" for the GPU.
Asmodian
8th October 2020, 01:15
I don't think there is anything significantly different in processing speeds but there are major differences in presentation so you might need to make changes to how presentation is setup.
I would make sure you are only using three or fewer frames presented in advance, Windows 10 usually doesn't seem to like more in my experiance. Also, FSE usually does more harm then good in Win10.
rack04
8th October 2020, 01:18
Check power profile in Nvidia control panel. Change it from "optimal" to "adaptive". Also make sure LAV decoder is set to "D3D11" and "Auto" for the GPU.
Appreciate the input. I changed the settings you described and the GPU is still balls to the wall trying to keep the queues as full as possible.
rack04
8th October 2020, 01:26
I don't think there is anything significantly different in processing speeds but there are major differences in presentation so you might need to make changes to how presentation is setup.
I would make sure you are only using three or fewer frames presented in advance, Windows 10 usually doesn't seem to like more in my experiance. Also, FSE usually does more harm then good in Win10.
Here is a picture of my general settings:
https://ibb.co/gdcFGHx
VBB
8th October 2020, 01:52
You want to uncheck:
- enable windowed overlay
- enable automatic full screen exclusive
- disable desktop composition
rack04
8th October 2020, 01:57
You want to uncheck:
- enable windowed overlay
- enable automatic full screen exclusive
- disable desktop composition
Thanks. I have unchecked those settings and it's still slower than Windows 7 but after lowering the quality settings from NGU Sharp (high) to NGU Sharp (Med) it's playable but the GPU is screaming. This likely just means that it's time to upgrade the GTX970.
VBB
8th October 2020, 02:09
What kind of content are you playing? The 970 lacks dedicated HEVC decoding.
rack04
8th October 2020, 02:16
What kind of content are you playing? The 970 lacks dedicated HEVC decoding.
1080p h264/vc1 Blu-rays @ 2560x1440 60Hz.
Asmodian
8th October 2020, 03:25
Thanks. I have unchecked those settings and it's still slower than Windows 7 but after lowering the quality settings from NGU Sharp (high) to NGU Sharp (Med) it's playable but the GPU is screaming. This likely just means that it's time to upgrade the GTX970.
One setting is not on that page:
https://hosting.photobucket.com/images/dd496/asmodian3/madVR_Windowed_Settings(1).png?width=590&height=370&fit=bounds
vBulletin® v3.8.11, Copyright ©2000-2026, vBulletin Solutions Inc.