madVR - high quality video renderer (GPU assisted)

[Sunset1982]

madshi, playback of my video files sometimes start paused and i have to resume it by hitting "space". It only happens sometimes... Any idea what error this could be?

[Dorohedoro]

Originally Posted by fluidz

Hi guys, i'm getting slight pauses in playback that last for approx 1-2 seconds, it happens once every few minutes.

Madvr 0.91.11 , Jriver Media Player 21, Windows 10, 1080ti - 385.41

This has only started to happen recently but I can't put my finger on whats causing it.

Your rendering time is too high and you are losing frames because of that, lower a little bit your settings and aim for 20ms~25ms... 40ms is too high.

[ForceX]

Originally Posted by madshi

For some reason the decoder queue seems to be struggling, too, which seems kinda weird if it runs smoothly in windowed mode. Also, rendering times seem very high, how are they in windowed mode?

Are you using the "device\display modes\hack Direct3D..." option in madVR? If so, does disabling that one help? Alternatively, you could also try to reduce the number of pre-presented frames and/or the size of the GPU queue, just in case you're running into VRAM trouble.

I did have Hack D3D option enabled but disabling didn't help since I'm running 60fps content to begin with. I have disabled that and tried it with shorter queues, but it wasn't the issue since the GTX 1080 has 8GB of RAM. Also, decoder queue shouldn't be struggling since the gpu can easily handle 4k60fps content.

Fullscreen with overlay element: http://i.imgur.com/TFeLoSb.jpg
Fullscreen: http://i.imgur.com/uoPS6ga.jpg

The rendering time behaves weirdly in fullscreen with max render time (5s) sometimes jumping to 100 ms, but in a window it always stays near average render time.

[YxP]

You have your player at adaptive or high performance power state in nvidia settings?

[Asmodian]

Originally Posted by 70MM

I read this...
It would be better to avoid the second processing step if possible, the Lumagen will mess up madVR's dithering (causing some loss in precision)

I will turn the dithering ON again in the Lumagen in conjunction with madvrs dithering, also ON.

Sorry, nevcairiel was correct. You have to finish that sentence: ", but it is still better to dither after processing than not."

What does the Lumagen Pro do that is beneficial?

Originally Posted by 70MM

Lumagen tells us that they process in 422 and to always use 422 out of the Lumagen regardless of what you input.

Ouch.

I assume this image shows 4:2:2 then? http://madshi.net/madVR/ChromaRes.png
At least you don't need to worry about madVR's chroma scaling options very much.

Edit: A Lumagen is good for people who are not using madVR but once you are using madVR a Lumagen only damages the video signal. madVR uses your GPU to do everything a Lumagen does but at higher precision and full resolution. When going for maximum quality 4:2:2 processing is completely unacceptable, down sampling the chroma to half the horizontal resolution for processing will lose a lot of quality.

[chros]

Originally Posted by hajosattila

I have calibrated my Tv display for gamma bt1886. (HCFR)
How should I configure madvr? Pure power curve 2.20 or 2.40 or BT.709/601 curve?

Hi Attila!

0. Here's the dedicated thread for calibration. And here's a small explanation by Asmodian wht the different options mean (but it won't give you much of a help with your question).

1. I'm not sure, but I think you have to go for "BT.709/601 curve" and one of the values (preferably 2.20) since there's no better option.
But are you sure that that's the best mode (bt1886) for your display and you?

2. But: what you did is the "old" way to do calibration. The better (final ) one is profiling: creating 3dlut with DisplayCal that can be used by madvr.
Feel free to ask more in the above mentioned calibration thread.

[Asmodian]

Originally Posted by hajosattila

I have calibrated my Tv display for gamma bt1886. (HCFR)
How should I configure madvr? Pure power curve 2.20 or 2.40 or BT.709/601 curve?

BT.1886 is pure power 2.4 if black = 0 nits, so if you have a pretty low black level use pure power 2.4. Never use the BT.709/601 curve unless you are trying to see shadow details on a bad display (and then you would set it in color & gamma, not calibration).

Note that these options don't do anything unless you start messing with gamma in "color & gamma", you are simply telling madVR the starting point in case you want to make adjustments.

And do check out DisplayCal, it isn't too hard to create a good 3DLUT for madVR now.

(Sorry, Cunningham's Law in effect again )

[Siso]

I'm wondering if image doubling is worth for 1080p files-> http://imgbox.com/XMGeHv9q

As for 720p it is ok -> http://imgbox.com/DATSnfxQ

So for 1080p files, I get bicubic 150 AR as the final step for luma?

[ForceX]

Originally Posted by YxP

You have your player at adaptive or high performance power state in nvidia settings?

I have it at Prefer maximum performance.

After getting frustrated with MPC-BE, I dusted out PotPlayer (blasphemy. I know) to give it a try and to my utter rage, it works fine. I now feel like an idiot for wasting so much time tinkering madVR on MPC-BE.

[Asmodian]

Doubling for 1080p -> 1440p has always been questionable. I do like doubling for 1.5x or more but for 1.33x simply using Jinc upscaling often looks almost identical for a lot less power. Bicubic 150 AR is downscaling it again, you can change it with the "<-- downscaling algo" option under "if any (more) scaling needs to be done:". SSIM 1D100 AR might look a bit better to you, or Bicubic60, but IMO nothing significantly beats Jinc for that upscale.

Do you not have the power to go up to NGU AA high? You have some room left in your rendering times.

[Siso]

Originally Posted by Asmodian

Doubling for 1080p -> 1440p has always been questionable. I do like doubling for 1.5x or more but for 1.33x simply using Jinc upscaling often looks almost identical for a lot less power. Bicubic 150 AR is downscaling it again, you can change it with the "<-- downscaling algo" option under "if any (more) scaling needs to be done:". SSIM 1D100 AR might look a bit better to you, or Bicubic60, but IMO nothing significantly beats Jinc for that upscale.

Do you not have the power to go up to NGU AA high? You have some room left in your rendering times.

For downscaling algo, it was left on "let madvr decide". I put SSIM 1D100 AR to see how it is.

What do you mean by "nothing beats Jinc for that upscale"?

[Asmodian]

I meant that for small upscaling ratios Jinc and NGUAA+SSIM look so similar that the improvement from spending all that GPU power is very small (Jinc is the best algorithm in madVR that can directly upscale to any size, not just 2x). 1.33x is on the line between worth it and not, where it is hard to decide.

[Siso]

Originally Posted by Asmodian

I meant that for small upscaling ratios Jinc and NGUAA+SSIM look so similar that the improvement from spending all that GPU power is very small (Jinc is the best algorithm in madVR that can directly upscale to any size, not just 2x). 1.33x is on the line between worth it and not, where it is hard to decide.

I see, good point! How about instead of Jinc for chroma, to put NG AA low?

[Shawns1234]

All, first time posting. I have kept up with this thread over the years and I have had an issue for some time now.

I have Windows 10 with CU. An AMD 480, madvr and Lab Filters are the latest available. I use a 4k Vizio m70. I am now using D3D11 in LAV now and it has been so much better, but I still have an issue where my PC locks up if I use NGU or any "doubling" algorithm. I can use xbr and cubic for chroma. The only option I can use for the other is cubic. Anything else, along with many other sharpening settings, and my PC locks up immediately after going full screen. I have tried every combo of settings, but this happens on my 480 and my 460 video card. If I use EVR, I do not have this issue. This has been going on for at least 6 months now. Any ideas?

[HillieSan]

I have no issues of this kind with my AMD RX 480. Perhaps you should check your power supply unit (PSU) if it provides enough power (Watts). The RX will consume lots of power when using NGU.

[Oguignant]

Hi. I have problems with the latest version (0.92.2) . When I want to watch a HDR movie in full screen exclusive mode, the Tv HDR mode is disabled. I already tried to make changes to the configuration, but I can not fix it. With the previous version (0.92.1) works perfect. Any ideas?

Thks!!!

[JayeDog]

Originally Posted by Dorohedoro

I will tell you my sharpening options tonight , I'm not in my house right now. Ffdshow is an old filter that will give you some problems (none for me so far) specially if you want to use hdr (I'm not using it yet).

Under the ffdshow options you will find some denoise filters, I really like denoise 3d because it will help you get rid of that ugly grain of some old movies (I dont don't like grain) without losing to much fine details. Until Madshi add some denoise options in madvr, ffdshow raw is a good option for me.

Hi, do you mind posting your denoise settings in ffdshoew raw?

[Asmodian]

Originally Posted by Siso

I see, good point! How about instead of Jinc for chroma, to put NG AA low?

You mean instead of Bicubic60AR? Your screen shot is using Bicubic60AR for chroma doubling.

NGU AA low is a good choice but any differences will be very minor, it would be better to put more power into luma, if you can.

[x7007]

I have a weird issue with 0.92.2 . when first entering the first movie at least after windows restart, the movie doesn't use the full upload queue , render queue and present queue. it only happen when entering D3D11 Full Screen 10bit . with D3D11 windowed 8 bit it works fine. not sure if it's the Lav Filters 0.70.2-74-git D3D11 issue.
after exiting and renetering it works fine. when entering the 2nd time to the same movie the Rendering time also reduced to 15-14 ms instead 26-24ms. I checked GPUZ and it shows full utilization of the core and mem .

EDIT: It also randomly happens while watching a movie , Decoder queue becomes 22 instead 24/24 . upload queue 5-8/8 render queue 3-8/8 and present queue is 0-4/4.
Do we must use Maximum performance on nvidia now for potplayer or MPC ? instead optimal power or adaptive ?

EDIT : It was something with the Cpu queue size , when it was 24 it happened , when I lowered it in real time to 16 it fixed itself .

hmm I didn't see this must rule "The rule is that the CPU Queue size must be bigger or equal to the GPU Queue size." that's why it's 16 and 8 ..


Potplayer is using Optimal power on the nvidia.
https://imgur.com/LzKlJay

[huhn]

rendertiems down matter much because modern GPU are changing the clock as they pleases so on low load the rendertimes can be really high.

optimal power settings can make problems when using madVR for me it is making major problem on my PCs adaptive is working fine for me.

some fluctuation in the queue are pretty meaningless as long as you are not dropping or repeating a frames for that reason.

you are using d3d11 which is not released yet and you ask question like:

Do we must use Maximum performance on nvidia now for potplayer or MPC ? instead optimal power or adaptive ?

so the first thing is to find out why. so in this case you should disable d3d11 decoding and check if it happens again.