madVR - high quality video renderer (GPU assisted)

[mytbyte]

What kind of content are you playing? The 970 lacks dedicated HEVC decoding.Exactly why I chose GTX960 at the time...

Sent from my Mi 9T using Tapatalk

[ND2560]

So maybe I missed it but is there any way/solution to properly use madvr&hdr&potplayer with the newest/recent nvidia drivers? Has anyone found a solution/tricks or do I have to stay put with my 446.14 driver. It's not that I have a problem with old drivers(for some games and firefox I prefer it) it's more the fact that when Cyberpunk eventually comes out I really like to have the newest drivers.

[Klaus1189]

Welcome to doom9, please read AMD, Intel and Nvidia driver issues and last recommended version (https://forum.doom9.org/showthread.php?p=1862365#post1862365)
Just use newest driver it should be OK. Nvidia is diligent now on HTPC stuff.

[ND2560]

Welcome to doom9, please read AMD, Intel and Nvidia driver issues and last recommended version (https://forum.doom9.org/showthread.php?p=1862365#post1862365)
Just use newest driver it should be OK. Nvidia is diligent now on HTPC stuff.

Thank you!!!!

[Siso]

I have an odd problem using jinc for image upscaling, in overlay mode the composition rate goes crazy it should be rock stable 120 hz, but it changes to 60hz, 40hz, 24hz. This only happens with jinc as upscaler, with any other upscaler everything is ok. Also in new path with frames in advance it is creating huge amount of presentation glitches.

[huhn]

overlay rendering shouldn't care about the composition rate so it shouldn't matter.

[Siso]

overlay rendering shouldn't care about the composition rate so it shouldn't matter.

That's good, but why does it make so many presentation glitches in new path mode? The card is GTX 1050 Ti.

[SirMaster]

You want to uncheck:

- enable windowed overlay
- enable automatic full screen exclusive
- disable desktop composition

Why disable windowed overlay?

[VBB]

So that he can use D3D11 instead of D3D9, for performance reasons.

[huhn]

d3d11 presentation isn't faster.
d3d11 native is always copied to d3d9 even with d3d11 presentation.

[VBB]

You're right. I got that confused with not being able to use 10-bit with D3D9 presentation, if so desired.

[scott1256ca]

I purchased a gtx 1650 super a couple of weeks back. I'm no expert in any of this, but I thought I'd post a bit about my success or lack of success in using it.
I also purchased a 4k hdr screen for use as a monitor.
In 4k @ 3840x2160 59.94 I can get 8 bit RGB full or ycbcr 444 limited. With no artifact removal I see < 50% GPU usage using 3G of memory.
If I reduce compression artifacts allowing NGU sharp even at low quality, I get dropped frames and about 70% GPU usage and pretty much all the 4G on the card
In 4k @3840x2160 23.976 I can get 12 bit RGB full or ycbcr 444 limited, but still 12 bit. With no artifact removal I see < 50% GPU usage using 3G of memory.
If I reduce compression artifacts allowing NGU sharp at low quality, I don't get dropped frames and about 70% GPU usage. At medium it won't even load the video

Everything says mapping hdr from bt.2020 to DCI-P3. I've tried changing the target gamut to BT.2020, but it still converts to DCI-P3. The tv can accept bt.2020. Is there a way to get it out?

[VBB]

Since we were talking about D3D9 and overlay earlier, I thought I'd run some tests. Turns out that HDR playback with anything D3D9 is broken, at least with the current driver. Everything stays washed out, regardless of FS mode. Enable overlay and the washing out gets even worse. YMMV.

[huhn]

overlay HDR only worked with 456.36 for me currently HDR doesn't work at all and i don't have an nvidia card connected to an HDR display anymore the ridiculously buggy rx 5700 XT is to blame in this case...

edit: scott1256ca check the power setting in the nvidia control panel.

the calibration settings should match the display setting if you are using send BT2020 you should leave the video at BT2020 even through the screen still has to lower this to DCI P3 or lower that'S unrelated.

[jkauff]

Everything says mapping hdr from bt.2020 to DCI-P3. I've tried changing the target gamut to BT.2020, but it still converts to DCI-P3. The tv can accept bt.2020. Is there a way to get it out?
Here's an excerpt from a Sound and Vision article:

"Another confusing issue with HDR is the BT.2020 color space that it's associated with. No current consumer video displays can provide full coverage of BT.2020, a color space that was designed to allow headroom for future technology developments. Movies are largely mastered in the P3 format, which is wider than the REC. 709 one used for HDTV, but not as wide as BT.2020. However, P3 is not a consumer format, and it's incompatible with the color matrix used for consumer video."

Your display may accept BT.2020, but it's going to convert it to DCI-P3.

[el Filou]

The arrow in "BT.2020 -> DCI-P3" in madVR's OSD doesn't mean it's downconverting to DCI-P3, it means mastered to DCI-P3 in a BT.2020 container.
I think it's misguiding IMHO.

[scott1256ca]

The arrow in "BT.2020 -> DCI-P3" in madVR's OSD doesn't mean it's downconverting to DCI-P3, it means mastered to DCI-P3 in a BT.2020 container.
I think it's misguiding IMHO.
ah ok, good to know.
According to rtings, my tv can only show about 60% of bt.2020 anyway, but does 90+% coverage of DCI-P3, so it wouldn't make much difference anyway. It wasn't something I was desperate to see, just interested.
Thx

[*MidnightWatcher*]

The arrow in "BT.2020 -> DCI-P3" in madVR's OSD doesn't mean it's downconverting to DCI-P3, it means mastered to DCI-P3 in a BT.2020 container.
I think it's misguiding IMHO.

But I believe that if your display is calibrated to BT2020 we need to ensure that madVR is set to BT2020, not to DCI-P3. When I had it set to DCI-P3 some of the colors were off slightly, but much better when set as BT2020.

[mytbyte]

Hmmm. You should not use BT.2020 if it's not being signaled to the display via "report BT.2020 to display". It is very inconclusive why BT.2020 may look better to you unless a measurement is performed. If you have a known Rec709 display side by side to the 4K HDR TV, and use "Rec709 in BT.2020" test patterns on the 4K HDR display, you can compare by eye and see what settings produce the closest match.

[leeperry]

I guess W10 is still a crapshot with mVR please?

Asking because I'd love to get UAC2 support but then again I want mVR to work equally well as it does in W7 :o

This said, sometimes I get black & white picture in FS when it was in color in windowed mode, also PotPlayer keeps making mVR unhappy with an error message on top left corner so how bad can it get with my HD7850?

[varekai]

Win7 x64
madVR+PotPlayer
Will there be any performance improvements placing madVR in same folder as PotPlayer?
C:\Program Files\DAUM\PotPlayer

[Morku]

Hi,

I need some help. I really tried to fix on my own, but I am stuck at an error. Most tutorials seems to be old and maybe not current. I am having repeated frames after changing my playback system.

Before, I had a system with Radeon™ RX Vega M GH graphics output. It was connected to my TV (single monitor). All I had to do was insert "1080p23, 1080p24, 1080p50, 1080p59, 1080p60" in display modes and all movies with 23.976, 24, 50, 29.970 etc. just played fine without any frame drops and repeat (or after hours after the movie end).

Now, I had to change my system with NVIDIA GTX 980Ti Output. The TV is connected as secondary monitor.
The TV is the same, the cables are the same, the videofiles are the same, all the configuration is the same.
MPC-HC is configured to display directly on the secondary monitor. ffdshow audio decoder is used to enable "AC3 (S/PDIF encode mode)" to have 5.1 AC3 Audio Passthrough with a S/PDIF connected cable.

Now, I have repeated frames every 12 minutes, which is of course noticable and distracting :(

OSD Screenshot after ~15 minutes playback (https://i.imgur.com/wYyny2E.png)

I have followed this tutorial: http://madvr.com/crt/CustomResTutorial.html

And get the following optimized pixel clock modes (https://i.imgur.com/Cq5BLnn.png)

As soon I Test mode, I get the "The GPU driver rejected mode, for unknown reasons". error :(
Yes, I have read comments that this is a "NVIDIA driver bug"... in 2017... now we have 2020...

Is there anything I can do? Maybe there are other, new tutorials? Or is this only fixable with new hardware? If so, is NVIDIA inappropriate as HTPC solution?

OS and drivers are latest: Windows 2004 Build 19041.546, NVIDIA 456.71 DCH

Thank you for help.

[Siso]

Win7 x64
madVR+PotPlayer
Will there be any performance improvements placing madVR in same folder as PotPlayer?
C:\Program Files\DAUM\PotPlayer

I think not.

[VBB]

I guess W10 is still a crapshot with mVR please?

What makes you think that? Most of us here have been using Windows 10 for a really long time now. Nothing to report on my end other than a few remaining driver issues, which are not deal-breaking at this point. I don't see any reason to stick with an eleven-year-old OS :)

[tp4tissue]

But I believe that if your display is calibrated to BT2020 we need to ensure that madVR is set to BT2020, not to DCI-P3. When I had it set to DCI-P3 some of the colors were off slightly, but much better when set as BT2020.

You need a colorimeter to figure out what space the display is using. It's kinda random out there, and some modes it clamps, but then if you enable something it unclamps, and then for whatever reason if you set that thing you enabled down 3 notchs, it clamps again. So it's a minefield.

But specific to 2020 vs p3, is the slightly different primaries. So it'll produce slightly different shades than intended.

You can't really correct this with the CMS on the TV, you need a tv that supports 3Dlut or you need to use Madvr's 3D LUT function.

IN PC mode, in most cases, it'll be something similar to P3 space, but the saturation will be pretty off.

[tp4tissue]

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.

We don't know FOR SURE what's happening.

Because there are 2 seperate machines, and they will both do whatever they want blind.

I'm reasonably sure the monitor you have is 8bit + frc dithering. So, it's -effectively- a 10 bit panel. a 10bit signal should be smoother.


Without worrying about about the signal chain, we can assess banding BY EYE.

Using the following test patterns. First post of this vvv thread. download both sets.

https://www.avsforum.com/threads/10-bit-gradient-test-patterns.2269338/


Turn the amd dithering on.

Then test the 4 conditions I mentioned in my other post.

Then Turn the amd dithering off.

Retest the 4 conditions

HDMI 2 can not support 3840x2160 @ 60hz 10bit, it can at 24hz.

Setting the madvr dithering toggle to a hot key makes testing easier.


Pick the one that's smoothest based on the Rotating pattern right side.

On my nvidia 1060 and 1660ti , going with 10bit pipe + dithering produces the smoothest gradient on my setup. This is the case for my Roku platform TV as well as my older u2410 10 bit dell monitor.

[PhantomF]

I just upgraded to an RTX 3080 from a 1070Ti on my HTPC connected with HDMI to my 4K Sony TV and am having a small issue. I use MPC-HC 1.9.7 with madvr and noticed after upgrading there was a slight jerkiness in my videos every few seconds that wasn't present when using the 1070. After experimenting I discovered that I need to set the LAV video decoder HW accel setting to use D3D11 and auto in order to get smooth playback. Not sure why but DXVA2 copy back was working just fine with the 1070.

Now that this is resolved another issue presented itself regarding the automatic display mode switcher in madvr. For some reason with D3D11 and auto selected as the decoder whenever I go full screen and it automatically changes the display to 2160p @ 23Hz the video zooms in to fill the whole screen and just freezes up, completely unresponsive to the playback controls even after coming out of full screen. If I manually change the display to 2160p @ 23Hz before I start the video it plays back just fine in full screen. Also the automatic full screen change function works great when any other decoder is selected in the LAV settings. Any ideas why it doesn't work with D3D11 mode or any fix available? Very annoying having to manually change the display mode before starting a video every time.

[leeperry]

What makes you think that? Most of us here have been using Windows 10 for a really long time now. Nothing to report on my end other than a few remaining driver issues, which are not deal-breaking at this point. I don't see any reason to stick with an eleven-year-old OS :)
madshi has stated very clearly on many occasions that W10 gave a number of issues with mVR, mostly related to broken D3D code IIRC and he strongly advised to stick to W8.1

[huhn]

that's not win 7...

[varekai]

I think not.
Thanks for the short and consize answer.

[varekai]

I don't see any reason to stick with an eleven-year-old OS :)I do...:D

[mytbyte]

We don't know FOR SURE what's happening.

Because there are 2 seperate machines, and they will both do whatever they want blind.

I'm reasonably sure the monitor you have is 8bit + frc dithering. So, it's -effectively- a 10 bit panel. a 10bit signal should be smoother.


Without worrying about about the signal chain, we can assess banding BY EYE.

Using the following test patterns. First post of this vvv thread. download both sets.

https://www.avsforum.com/threads/10-bit-gradient-test-patterns.2269338/


Turn the amd dithering on.

Then test the 4 conditions I mentioned in my other post.

Then Turn the amd dithering off.

Retest the 4 conditions

HDMI 2 can not support 3840x2160 @ 60hz 10bit, it can at 24hz.

Setting the madvr dithering toggle to a hot key makes testing easier.


Pick the one that's smoothest based on the Rotating pattern right side.

On my nvidia 1060 and 1660ti , going with 10bit pipe + dithering produces the smoothest gradient on my setup. This is the case for my Roku platform TV as well as my older u2410 10 bit dell monitor.

I currently keep DP_Disabledither=1 because a kind forum member pointed out to me the CRU app through which I managed to summon 10 and 12 bit options for the monitor. It is 8bit+FRC.

I do have GPU's gamma ramps enabled because I calibrated via DisplayCAL to gamma 2.2 from monitor's native 2.26.

I cannot tell by eye whether AMD 10-bit + madVR 10-bit with errordiff. dithering produce better gradients than if MadVR dithering=none, but if I set an all-8-bit pipeline AMD + MadVR + MadVR's dither=error diffusion, I get colored vertical strips here and there with gamma ramps enabled. If I disable gamma ramps (i.e revert to monitor's native gamma curve) or use an all-10 bit pipeline, the colored stripes are not readily visible or actually not visible at. However even with all-10 bit pipeline there is a kind of slightly visible transient at around 15% IRE due to the gamma ramp tweak.

Overall, I'm now a happy puppy, albeit not entirely since I cannot make use of monitor's 10 bit capability in OpenGL (Photoshop, After Effects, possibly Resolve).

[huhn]

the only reason you should disable dithering is to prove it should not be disabled. this option has nothing to do with the screen just with internal calculations.

[mytbyte]

the only reason you should disable dithering is to prove it should not be disabled. this option has nothing to do with the screen just with internal calculations.

AMD dithering o MadVR dithering?

[tp4tissue]

I currently keep DP_Disabledither=1 because a kind forum member pointed out to me the CRU app through which I managed to summon 10 and 12 bit options for the monitor. It is 8bit+FRC.

I do have GPU's gamma ramps enabled because I calibrated via DisplayCAL to gamma 2.2 from monitor's native 2.26.

I cannot tell by eye whether AMD 10-bit + madVR 10-bit with errordiff. dithering produce better gradients than if MadVR dithering=none, but if I set an all-8-bit pipeline AMD + MadVR + MadVR's dither=error diffusion, I get colored vertical strips here and there with gamma ramps enabled. If I disable gamma ramps (i.e revert to monitor's native gamma curve) or use an all-10 bit pipeline, the colored stripes are not readily visible or actually not visible at. However even with all-10 bit pipeline there is a kind of slightly visible transient at around 15% IRE due to the gamma ramp tweak.

Overall, I'm now a happy puppy, albeit not entirely since I cannot make use of monitor's 10 bit capability in OpenGL (Photoshop, After Effects, possibly Resolve).


Use the 3Dlut w/embedded vcgt with madvr instead of the gamma ramp.

The dispcal loader will automatically disable the gamma ramp and let madvr take the entire wheel . This is the only way to ensure color accuracy through madvr.

You should not be using the 1dlut for madvr

That said, the lut doesn't impact the dithering smoothness.

You can see the smoothness difference between the modes better if you zoom in to the dark area.

[tp4tissue]

the only reason you should disable dithering is to prove it should not be disabled. this option has nothing to do with the screen just with internal calculations.

disabling it is useful for testing the screen's default behavior with 10/8bit input signals.

[VBB]

madshi has stated very clearly on many occasions that W10 gave a number of issues with mVR, mostly related to broken D3D code IIRC and he strongly advised to stick to W8.1

Keep in mind, it's been years since he last commented on that. Windows 10 is more than five years old at this point. I'm not saying there aren't reasons to stick with an older OS, but madVR is not one of them.

[mytbyte]

Use the 3Dlut w/embedded vcgt with madvr instead of the gamma ramp.

The dispcal loader will automatically disable the gamma ramp and let madvr take the entire wheel . This is the only way to ensure color accuracy through madvr.

You should not be using the 1dlut for madvr

That said, the lut doesn't impact the dithering smoothness.

You can see the smoothness difference between the modes better if you zoom in to the dark area.

I thought it a good idea to have a gamma 2.2 applied system-wide via simply letting the profile loader modify the gamma curves but I will try making a specific MadVR LUT with vcgt burned-in.

I found out DisplayCAL Profile loader will detect MadVR, but will unload the profile and reset the gamma curves only if set so in MadVR.

[tp4tissue]

I thought it a good idea to have a gamma 2.2 applied system-wide via simply letting the profile loader modify the gamma curves but I will try making a specific MadVR LUT with vcgt burned-in.

I found out DisplayCAL Profile loader will detect MadVR, but will unload the profile and reset the gamma curves only if set so in MadVR.

it's automatic. if you enable the 3dlut with the vcgt baked in, the loader will disable it automatically.

I recommend you set it to a custom tone curve, 2.4 relative, otuput offset 100%.Set it in the calibration tab as well as the LUT tab. use <slow>

I've come up against bugs where the white point doesn't apply unless you set the tone curve in the calibration tab.

If you're using an ips panel, I highly recommend you do pre-testing to find the mode that matches 2.2gamma and set the rgb manual to maximize contrast. Do not use bt1886 with ips, it will look terribad. Also pre-test the contrast ratio/ white point in the various monitor modes.

The less correction madvr has to do, the better.

IPS, you'll probably want the panel at 60 to 100 nits, watched in the dark. If you push 200 nits, the floaty blacks of ips will blow out shadow detail.

HDR, you don't have a choice, set everything to max but don't touch the base rgb settings after you setup for sdr.

Then run a seperate 3dlut loop again for the 250nit setting. Select source color space rec2020 and p3, adjust for 2.2 gamma instead because Madvr assumes 2.2 gamma during tonemapping. deselect create 3dlut after profiling, you can use the same profile session to create as many 3dluts as you need. But you'll want to set the custom gamma in the calibration tab in all cases during initial profiling.

The thing I've never actually been sure about is whether Madvr's patch generator will go into 10bit exclusive or use the 10bit pipe. But it shouldn't affect the color performance that much. Either way, you'd want to put this in full screen disable the osd in the Madtpg

[tp4tissue]

Does anyone know if Madtpg will use 10bit mode when fullscreen / go exclusive ?

[chros]

I recommend you set it to a custom tone curve, 2.4 relative, otuput offset 100%.Set it in the calibration tab as well as the LUT tab. use <slow>

I've come up against bugs where the white point doesn't apply unless you set the tone curve in the calibration tab.
What do you mean about the White Point issue?
I have/had the problem (https://www.avsforum.com/threads/2018-lg-oled-calibration-and-user-settings-no-price-talk.2962814/post-59016662) where WP gets lower (resulting in warmer image) than desired.
I guess I'll give it another try.

Does anyone know if Madtpg will use 10bit mode when fullscreen / go exclusive ?
Excellent question, I'm not sure, but I guess so if it works that way in a normal window.

[tp4tissue]

What do you mean about the White Point issue?
I have/had the problem (https://www.avsforum.com/threads/2018-lg-oled-calibration-and-user-settings-no-price-talk.2962814/post-59016662) where WP gets lower (resulting in warmer image) than desired.
I guess I'll give it another try.


Excellent question, I'm not sure, but I guess so if it works that way in a normal window.


For Oled, anything related to white is probably ABL .

Specific to what I was refering to is that if you don't set a custom tonecurve in the calibration tab, the 3dlut sometimes doesn't adapt to the white point selected <in the calibration tab>

[huhn]

madTPG uses your madVR setting if you setup 10 bit it's supposed to use 10 bit.

[mytbyte]

If you're using an ips panel, I highly recommend you do pre-testing to find the mode that matches 2.2gamma and set the rgb manual to maximize contrast. Do not use bt1886 with ips, it will look terribad. Also pre-test the contrast ratio/ white point in the various monitor modes.

The less correction madvr has to do, the better.

IPS, you'll probably want the panel at 60 to 100 nits, watched in the dark. If you push 200 nits, the floaty blacks of ips will blow out shadow detail.



All monitor's picture mode presets return the native gamma of 2.25...that's not a disaster but I'd like to have 2.2 flat for productivity work. The contrast is very good for an IPS, cca 1400:1 at 100 nits. WB is dE 0.6 peak at 10% white after white point calibration in custom mode (since all modes have the same gamma anyway).

I'll continue to play as you suggested and come back with more questions if needed.

[Asmodian]

madTPG uses your madVR setting if you setup 10 bit it's supposed to use 10 bit.

I believe it uses the same .ax, it doesn't have its own present logic, so works the same way naturally.

I did some testing a long time ago, when 10 bit was still pretty new, and it did seem to change to 10 bit in 10 bit mode (dithering off for testing). I didn't feel it was totally conclusive but full 10 bit throughout the chain always feels a little bit uncertain. :o

I thought it a good idea to have a gamma 2.2 applied system-wide via simply letting the profile loader modify the gamma curves but I will try making a specific MadVR LUT with vcgt burned-in.

I also recommend embedding the same 1DLUTs into madVR's 3DLUTs as you normally have loaded, better all around with no downsides I know of.

[chros]

For Oled, anything related to white is probably ABL
ABL is not an issue on Oled either (you can disbale it/them in Service Menu), especially with low nits SDR (100-150nits).

[mclingo]

oddly, ABL is called local dimming on my 2015 OLED, I checked it again yesterday to make sure it was off, with it on I would get constant full page dimming in web pages, nothing in movies though.

[Andy o]

So I was surprised to learn that mpv has been supporting G-Sync for quite a while now. I installed it and sure enough, it works on my 65C9. It seamlessly switches refresh rate without blanking the screen, basically what HDMI 2.1's QMS is supposed to do. There's even an old post (https://forums.blurbusters.com/viewtopic.php?t=3509) (from 2017!) at blurbusters saying it works with VFR video. Has there been anything on madVR supporting it since VRR got more mainstream with the newer TVs? I can't find anything about it, but I remember years old posts asking for it and madshi (IIRC) saying it wasn't feasible at the time.

[ND2560]

Can someone help me with potplayer&madvr. When I watch movies with HDR the picture looks too dark somehow. Like not by much.
Like for example I watched a Batman Returns 4k scene on youtube with HDR on(in windows settings) and it looked brighter than my file on potplayer&madvr. Like on youtube you could see the texture on batman cape and sharp detailed hair on Michael Keaton but on my player it has big dark tones where these details should be.
With the potplayer control panel I can't adjust the video settings like birightness, contrast, etc. It doesn't affect the picture with madvr active.
Is there somewhere a setting on the madvr settings/or potplayer settings where I can bright the picture up? And by how much?

Thank you!

[el Filou]

@Andy o that's fantastic news, and I hope that's the future of media playback on any platform including the PC. This is also the ideal way of finally getting rid of dropped/repeated frames in any scenario and with any config.
I am very interested in that feature but haven't seen any news either on that front in madVR since the last time madshi talked about G-Sync.
I hope with VRR becoming mainstream the Envy will adopt it too and we will be able to test that in madVR. Ideally at one point there should be a standard API in Windows that any foreground app can use to dynamically and seamlessly match the display refresh to its content and we can run VRR all the time.