madVR - high quality video renderer (GPU assisted)

[Warner306]

Originally Posted by tp4tissue

Considering a used 1070 or 1080 right now.

Is the additional power much greater than a 1060, (FOR MADVR, I don't play games.)

I already own a 1060 for the kitchen pc.

1060 priced ~$160 used
1070 ~$250
1080 ~$400


What extra performance setting (MADVR) can be reached by the 1070/ 1080..

I would only bother if you are using the HDR -> SDR tone mapping. A 1070 or better can help there.

[tp4tissue]

Originally Posted by Warner306

I would only bother if you are using the HDR -> SDR tone mapping. A 1070 or better can help there.

What if it's HDR tonemap to HDR output on a FakeDR TV ?

Does a 1070 help with that ?

[Warner306]

Do have all of the trade quality for performance options related to tone mapping unchecked and highlight recovery strength enabled? If yes, then the GTX 1060 can still do the job (I would guess barely). If not, a GTX 1070 or better would help.

[griffind]

I need help evolving from 'poking at settings' to 'knows what poking at settings does'

Been using MadVR for a while now for HDR>SDR tone mapping on my Projector and I'm wondering if I am missing out on picture quality because I don't understand the settings properly.

Windows 10 - 64-bit
i7 - 16gb ram - NVIDIA 1070ti
madVR v0.92.17
LAV - 0.70.2.81-git

- My Calibrated Screen/PJ combo is hitting approx 70 nits with a rec.709.3dlut for Gamma 2.2 and I have this chosen in the calibration section of MadVR. I don't have any other 3DLut files generated.

- For HDR>SDR I am using shaders with a target peak of 200 nits, BT.2390 curve, balanced color tweaks and high highlight recovery.
I have 'measure each frame's peak luminance' checked but not sure if it works with NVIDIA CUVID decoding.

I wondering about how much I 'dont know' and if I'm getting less than stellar results with this config.

I 'think' my picture looks good, but without having any other point of reference I feel a bit in the dark.

- For my 4k HDR BT2020 content, should I be using a different LUT?
- Should I have Gamma processing enabled?
- How do I know if I am crushing blacks or if my highlights are right?
- How do I enable dynamic peak luminance adjustments?
- Why does my cat drink out of the toilet not her water bowl?

So many questions.

[Soxbrother]

Hello,

I've installed MadVR (latest version)
and this is my system :

HP Pavilion 300-030NB
Windows 10 - 64-bit
i3 - 4gb ram - intel HD 4400 graphics
Latest mediaportal
Latest LAV codec - set to dvxa2 copy back

Connected to my Samsung UEJS9000 tv ( maximum refresh rate 30hz ( according to the "windows settings page" )

I had setup MadVR, according to what different users of an intel HD graphics 4000 (I have 4400) said, which was to use dvxa2 settings and such.

But nothing seemed to prevent my video from not playing smooth/stuttering.

I've tried numerous settings and different codecs ( Lav, CoreAvc Pro ), but nothing seemed to give me smooth playback.

Until I tried the setting under "Devices - Generic PnP Monitor - display modes"
"list all display modes madVR may switch to"

The default setting had nothing in it, so I've now set it to "1080p23" and it worked.

When I try to use 2160p23, it gives a black screen with audio only ( Lav codec - dvxa2 copy-back )
I guess the 2160p23 gives the same result as when you leave the field blank.

So I'll use 1080p23 for now.

But that leaves me thinking, now when a video is played, it will play in 1080p on a 2160p screen.
So now it's not upscaling anything, right ?

So is there any difference in quality now with this setting to "1080p23"
between MadVR and the EVR renderer option in mediaportal settings ?

I also had almost everything checked on the settings page of "Trade quality for performance".

I'm going to uncheck one at a time now and see if quality can improve without performance.

Is there any way to upscale my 1080p video's to 4k resolution with this intel 4400 HD Graphics and have smooth video playback?
Or am I stuck with the 1080p23 setting for now ?

Thanks in advance.

Greetings,
Soxbrother

[tp4tissue]

Originally Posted by Soxbrother

Hello,

I've installed MadVR (latest version)
and this is my system :

HP Pavilion 300-030NB
Windows 10 - 64-bit
i3 - 4gb ram - intel HD 4400 graphics
Latest mediaportal
Latest LAV codec - set to dvxa2 copy back



Soxbrother

HD 4400 benches 567 on passmark.

You will have to turn off many of the madvr's features in order to scale 1920x1080 into 3840x2160.

Madvr will still give you more color accuracy, but to use higher quality filters like Lanczos, there's simply not enough processing power to do it.

If you scale 1080p to 1080p, then that is chroma scaling only, 1:1 luma. the 4400 will do this perfectly fine. You can enable Jinc or Lanczos for chroma.


I believe you also need displayport to hdmi 2.0 adapter, because 4400's native hdmi is probably not HDMI 2.0, which is required for 3840x2160 60hz

Also, run CPU decode on Lav, do not use dxva, because you want to save as much GPU time as possible for Madvr.

Letting the CPU decode the first part is fine.

[Asmodian]

Originally Posted by griffind

- For my 4k HDR BT2020 content, should I be using a different LUT?

Unneeded, especially since your projector is SDR.

Originally Posted by griffind

- Should I have Gamma processing enabled

No

Originally Posted by griffind

- How do I know if I am crushing blacks or if my highlights are right?

Use the AVS HD 709 Test patterns

Originally Posted by griffind

- How do I enable dynamic peak luminance adjustments?

Check the "measure each frame's peak luminance" box in the hdr settings page.

I would also recommend not using CUDA acceleration anymore, DXVA2 copy-back or D3D11 copy-back is just as good or better and as universal solutions they are getting more development.

[tp4tissue]

So... if there are large patches of very bright data, using highlight recovery, this can cause clipping right ?

I noticed some clipping on the helmets in film -interstellar- and near the end black hole escape scene.

[chros]

Originally Posted by tp4tissue

Also, run CPU decode on Lav, do not use dxva, because you want to save as much GPU time as possible for Madvr.

This is not how GPU hardware acceleration works: it's a completely separate pipeline. Just use GPU hardware acceleration all the time if you can.

[Axelpowa]

Hi guys,

I'm thinking about replacing my gtx1080 with a Rtx2070 which could be quite affordable money wise.

The reason is I use my gpu almost to the limit with all the hdr new stuff and with the "reduce random noise" filter which I enjoy a lot, but uses lots of resources.

Would it be a smart move or better get a gtx1080TI?

Regards!

[nevcairiel]

A 1080 and a 2070 are quite comparable in performance, that wouldn't be a huge "upgrade" to speak of. If you can get a cheap used 1080Ti, that might be a better idea if you really need more performance.

[tp4tissue]

Originally Posted by chros

This is not how GPU hardware acceleration works: it's a completely separate pipeline. Just use GPU hardware acceleration all the time if you can.

I'm not sure how the pipeline works, I'll take your word for it.

But in my own testing, if I use Dxva or Dx11, the render time is always slightly higher than if I use CPU decode.

This was especially true on my iGPU and slower graphics cards.

On my HD3000 for example it will stall out /drop frames if I use Dxva decode, whereas if i let it Cpu decode, and gpu render, it will play smoothly.

[Axelpowa]

Originally Posted by nevcairiel

A 1080 and a 2070 are quite comparable in performance, that wouldn't be a huge "upgrade" to speak of. If you can get a cheap used 1080Ti, that might be a better idea if you really need more performance.

Thx!!!

[chros]

Originally Posted by tp4tissue

This was especially true on my iGPU ...

I'm not sure about iGPUs (never used them for madVR, I even used the discrete GPU in an optimus system): whether they have small amount of dedicated memory or they use the system wide RAM.

[Alexkral]

Originally Posted by Asmodian

I would also recommend not using CUDA acceleration anymore, DXVA2 copy-back or D3D11 copy-back is just as good or better and as universal solutions they are getting more development.

I keep seeing this, and while I'm sure it's true for most cases I don't think it is for some. At least in my system (Windows 7, GTX 1080) I get framedrops with DXVA and D3D11 when playing 4k and up @ 60 fps, while CUVID runs smooth. This means that at least for this content CUVID is faster (it was even faster with old drivers but they produced some ugly artifacts with HDR). I don't know if this could be related with some particular rendering or scaling setting, but I don't think so because it's the same with other renderers and players. Maybe for VR in Windows 10 this is solved with that ATW thing.

[chros]

Originally Posted by Axelpowa

I'm thinking about replacing my gtx1080 with a Rtx2070 which could be quite affordable money wise.

According to the specs, a 1080 can be faster than a 2070
https://youtu.be/CT2o_FpNM4g?t=916

[nevcairiel]

Originally Posted by Alexkral

This means that at least for this content CUVID is faster (it was even faster with old drivers but they produced some ugly artifacts with HDR). I don't know if this could be related with some particular rendering or scaling setting, but I don't think so because it's the same with other renderers and players. Maybe for VR in Windows 10 this is solved with that ATW thing.

Since you mention HDR, you should know that HDR playback with CUVID will be much worse (in quality) then with any other mode, since CUVID does not actually export all the necessary metadata for full HDR reproduction.
D3D11 in Native mode should be the best you can do for performance (ie. without selecting a graphics device in LAV Video).

[huhn]

cuvid forces your GPU into high performance mode while dxva and d3d11 use the NCP power setting which is by default optimal and which is very well known to be problematic with madVR.

no matter which hardware decoder you select on nvidia you use the same decoder in the end the only thing that changes is how this data is transferred.

so go to the NCP and change the power setting to adaptive or maximum performance and try again.

[Warner306]

Originally Posted by griffind

I need help evolving from 'poking at settings' to 'knows what poking at settings does'

Been using MadVR for a while now for HDR>SDR tone mapping on my Projector and I'm wondering if I am missing out on picture quality because I don't understand the settings properly.

- How do I know if I am crushing blacks or if my highlights are right

Try using these HDR10 black clipping patterns to test black clipping (white clipping is not overly important):

https://drive.google.com/file/d/1PYY...8JqyuGH7-/view

The 8-bit/10-bit combined black clipping video should be paused when making adjustments because it is short. The 8-bit values are listed along the bottom of the pattern as thick bars. The faint, grey gradient should extend as close as possible to the edge of Bar 16 from right-to-left.

The target nits is tradeoff between brightness and contrast. Increasing the target nits will give you more highlight detail, but the entire image will become progressively darker to create necessary contrast.

The other options have been simplified in the latest test build. Hopefully, it comes out soon because I find it is hard to get a bad picture with the latest build.

[chros]

Originally Posted by huhn

so go to the NCP and change the power setting to adaptive or maximum performance and try again.

That's what I thought as well, until 2 weeks ago I had to set Optimal back in NCP because I had massive drops with Adaptive. Now it seem to work fine. (Setup is in my signature.)