madVR - high quality video renderer (GPU assisted)

[CarlosCaco]

No, I didn't change anything in that area from .22 to .23.

so in .22 it should be working with simple upscaling without (supersampling) 1280x720 to 1366x768?
in this case not working :( only working with supersampling...

[ShiftyFella]

ShiftyFella, performance of your 290x is amazing! I really hope that RX 480 will perform as well but i doubt it after seeing Sunset1982 test. He compared RX 480 to GTX 970 and they are pretty close in terms of performance. If i remember correctly, back in the days NNEDI3 worked faster on AMD cards and my 7870 were faster than 970. Maybe things changed since then but this is really concerns me.
I don't have 4k screen, otherwise I would do his test to compare numbers but at 1440p and setting I mentioned earlier, render time difference is about 1-2ms and sometimes less as you don't have to do downscaling from doubled 720p image

Maybe banding was the wrong word, but the sun looks odd to me with 100% luminance reduction
I think there is no perfect settings to this clip, looks at this scene:
100% luminance 0% saturation: http://i.imgur.com/GP9KSYC.jpg
0% luminance 100% saturation: http://i.imgur.com/6e5DiMB.jpg

either setting produces wrong clouds

The encode has values up to 10,000 nits which is clearly wrong. madVR has no idea how to handle this situation since it has no idea what the values should be. It makes no sense to tune an algorithm to faulty input data.
I guess you're right, at least sony demo stays withing 4000 nits. too bad there are so few hdr demos that you can play with madvr that are publicly available

[Uoppi]

There's one more thing I might have changed: In older versions when there was image doubling -> downscaling, IIRC I ran upscaling refinement on the downscaled final res. I might have changed that to run on the doubled image before downscaling. Which of course would also increase render times, but should also improve image quality.

Seems like this is what was causing the higher rendering times on versions after .20, at least in my case that is.

Any chance the old upscale refinement procedure could return as a trade quality for performance option (if there's a meaningful reason for that PQ-wise in the first place)?

Interestingly, for 720p -> 1080p upscales with SR2, I'm getting identical (practically to the millisecond) processing times for:

a) SSIM 2D on v.20
b) SSIM 1D on versions after .20

I currently lack the time to do any meaningful comparisons, but I'd be curious to know theoretically at least which one should provide the better PQ. I've settled on SSIM 1D for now because it's a bit more efficient than Jinc.

[zoyd]

Maybe banding was the wrong word, but the sun looks odd to me with 100% luminance reduction.

yes, I see what you mean and after checking it here there is some sort of artifact @100%, 0% or 25% looked ok.

This scene looked best at 50/50, so I think this method is too scene dependent.

lantern (https://dl.dropboxusercontent.com/u/2621383/sony2.png)

[x7007]

Can you please make a screenshot of the Ctrl+J OSD with these "worst quality" settings where the render queue doesn't fill in windowed mode?


I'll make as soon as I'm at work

It's weird. When I'm in Windowed mode and focused on the potplayer window it lowers the Queues . When I'm not in focus it goes back to full .... LoL :rolleyes:

Every time I click outside the video on desktop or something else and back on the video, the Queue changes and goes back again. After I did it couple of times the Queue are actually fine now. but it still changes to low - bad to high - good if I keep clicking like before.

https://s32.postimg.org/szqva7a81/Focused.jpg (https://postimg.org/image/szqva7a81/)

https://s32.postimg.org/96erhhwu9/No_focus.jpg (https://postimg.org/image/96erhhwu9/)

[ryrynz]

To my eyes camp looked better with 100% luminance

From my results with a Sony X8500D using the Cinema Pro preset I'd agree with this but maybe as madshi said we should be trying some other mixed options to find something that works well for most content.

[James Freeman]

The chess demo is encoded incorrectly- I wouldn't use it for eval.
I agree, madVR relies on the metadata and selected peak luminance value.
This Chess demo unfortunately is not a good source for testing.
Out of the two free demos we have, both are a total mess in terms of encoding and metadata. :rolleyes:

[K2mil]

Don't think anyone else does. Is it FSE mode? Are you using D3D11? Windowed overlay? Are you changing refresh rates? You could always just stay in windowed mode. Did you do a fresh install of your drivers of did you just install the 1070 straight in after taking out another Nvidia card?



Thanks I didn't have D3D11 checked now It seams to work perfect now. Yes I always do clean installation of drivers.

Any settings suggestion you would shear I'm very new to this and there is over 1000 pages :eek:

I should add that I'm on 1440p 144hz mostly playing mp4 1080p

[QBhd]

I have a couple of observations to report. First, the "ShowRenderSteps" trick is very cool, would be nice if the list continued in a second column so we can see everything, but it is what it is. Second, the "ShowRenderSteps" is actually more demanding than regular OSD... pushes my 720p profile way over the edge, more than likely due the many more lines of info that need to be drawn, again it is what it is and not a big deal.

Now to the important info to report, My setup: 1280x720 ==> 1024x768

1) Upscaling Refinements definitely were not being used in version 0.90.20... I had never noticed since I never really tested them, I had just copied a previous profile where upscaling refinement was actually being used when I created the 720p profiles. So it is good to know that what I thought was a spectacular image had nothing to do with US Refinements.

2) SSIM Downscaling is NOT being used in version 0.90.23 and this is very noticeable, Bicubic150 AR is OK, but I truly do prefer SSIM2D-100 w/AR-Relaxed

Here are some screen shots:

v0.90.23
http://image.prntscr.com/image/84ed565c630d4b4fa01634016a4ea81a.png


v0.90.20
http://image.prntscr.com/image/f8d1b57b883b4d5a9784d81446d72fb9.png


So I am going to stick with v0.90.20, but not for the reason I had originally thought (the "performance" hit)... broken downscaling is the real reason to stick with the older version.

QB

[leeperry]

I truly do prefer SSIM2D-100 w/AR-Relaxed
IIRC the relaxed/strict subsetting for downscaling AR doesn't do anything with SSIM?

[QBhd]

IIRC the relaxed/strict subsetting for downscaling AR doesn't do anything with SSIM?

Hmmm, I thought I had tested that and noticed a difference and stuck with the relaxed setting... either way, it's soft or relaxed or NO AR... I'll stick with AR on.

QB

[leeperry]

Hmmm, I thought I had tested that and noticed a difference and stuck with the relaxed setting... either way, it's soft or relaxed or NO AR... I'll stick with AR on.

I hear you, it all get pretty confusing at times.....anyway if SR AB is gone for good, I'll try my luck adding AB at the downscaling stage on top of SSIM 2D 100%. I forgot it's even there as I can't read it at all (http://www.imagebam.com/image/07cc3b495626515)(using big fonts on W7).

[QBhd]

Addition to my report from above:

SSIM downscaling in v0.90.23 does work for a full downscale (i.e. 1920x1080 ==> 1024x768)... last night I had forgot to check that scenario. So again it seems that mixed scaling is to blame :(

QB

[Mano]

Hi madshi, are you still using AMD card? If yes what driver version are you using?

[Warner306]

Thanks I didn't have D3D11 checked now It seams to work perfect now. Yes I always do clean installation of drivers.

Any settings suggestion you would shear I'm very new to this and there is over 1000 pages :eek:

I should add that I'm on 1440p 144hz mostly playing mp4 1080p

Try the guide in my signature.

[Georgel]

I've been using madVR with no Dithering at all for the past two months. It's better this way with anime.

When dithering is active, the background moves around a lot and is distracting, it gives a sense of flickering / a sense that there is something wrong with the screen.

At any rate, the dithering algorithms are great by themselves, it's just not that great with anime.

In other order of thoughts, seems the newest version of madVR works great for me!

[Sunset1982]

I ordered a GTX1060 today. Will add it to my comparison end of this week...

[pose]

Georgel, what setting do you use for 720p anime? Sunset1982 i will be waiting!

[Asmodian]

I've been using madVR with no Dithering at all for the past two months. It's better this way with anime.

When dithering is active, the background moves around a lot and is distracting, it gives a sense of flickering / a sense that there is something wrong with the screen.

At any rate, the dithering algorithms are great by themselves, it's just not that great with anime.

In other order of thoughts, seems the newest version of madVR works great for me!

You do not need to turn off dithering! To keep the background still simply turn off "change dither for every frame".

Dithering is very important for anime! Doesn't the banding bother you without it?

[pose]

Asmodian, can i also ask you which settings you prefer for anime?

[Asmodian]

Asmodian, can i also ask you which settings you prefer for anime?

For 720p24 to 2560x1440@144Hz I am currently using:
Chroma Upscaling: Reconstruction SharpAR + SR1
Luma Doubling: NNEDI3 256
Chroma Doubling: NNEDI3 32
Image Upscaling (not used): JincAR
Thin Edges: 0.6
Smooth Motion: On
Dithering: ED1, colored noise OFF, change dither OFF

Ordered dithering is also good. I have all artifact removal options disabled for this one, I will turn them on if the source needs it. I also play with the upscaling refinement options for different sources, I have never found a good set of 'default' settings for the refinement options. No trade quality for performance options checked.

[pose]

Asmodian, what card allows you to use so much neurons?

[seiyafan]

I ordered a GTX1060 today. Will add it to my comparison end of this week...

Thanks!

[ryrynz]

Asmodian, what card allows you to use so much neurons?

He has a 980 Ti.

[AntonP]

I have 2 problems with 3D playback using LAV + madvr.
Config i7-2600k + gtx980 + win 8.1, latest nightly builds of MPC-HC and LAV.
1. nvidia refresh rate problem - 23.971 instead of 23.976. in 2D mode i fixed it with 24Hz monitor refresh rate (24p instead of 23p in madvr settings) + reclock
in 3D i cant set 24Hz in madvr - 23p mode switches on automaticaly.... how can i set 24p in 3D?
2. artefacts in 3D mode in some scenes (every time in same moment). red car on screenshot is example of it. in powerdvd 16 all is ok, so its not BD3D problem. I think its decoder's problem, but not sure. how to fix this bug?

[chros]

I ordered a GTX1060 today. Will add it to my comparison end of this week...
Thanks!

[Warner306]

Thanks!
I'd like to know if the GTX 1060 is capable of 1080p -> 2160p with these settings:

Chroma: super-xbr100 + AR
Luma Doubling: NNEDI3 32 neurons
Image: Jinc + AR
Artifact Removal - Debanding On
Upscaling Refinement: SuperRes (3)
Dithering: Error Diffusion 2

[aufkrawall]

Most certainly it is impossible unless you remove Error Diffusion from that list (and probably even without it).

Edit: Ok, with <= 30fps it might work.

[huhn]

i don't think error diffusion is a good idea at UHD

[XTrojan]

i don't think error diffusion is a good idea at UHD

Care to explain? Is there a major drawback except performance hit?

Is there a major difference with error diffusion from 1080->2160p?

[Sunset1982]

I'd like to know if the GTX 1060 is capable of 1080p -> 2160p with these settings:

Chroma: super-xbr100 + AR
Luma Doubling: NNEDI3 32 neurons
Image: Jinc + AR
Artifact Removal - Debanding On
Upscaling Refinement: SuperRes (3)
Dithering: Error Diffusion 2

ok, I will test it. Card will arrive today



Edit:

Tested for you. File was 1920x1080 (with black bars), 23,976, 8 Bit AVC, D3D11 FSE, 10Bit mode:

27,04 ms / 140 w (with automatically detect hard coded black bars feature & crop black bars)
33,45 ms / 170 w (without detecting hard coded black bars)


tested with 1920x1080 (Avatar Mkv, real 16.9 fullscreen without black bars), 23,976, 8 Bit AVC:

34,1 ms / 172 w

[Sunset1982]

guys, my gtx1060 arrvied. Started my testings but I have a problem:

Somone knwos how to deactivate the different powerstates of this card?

gtx1070's performance looked stable and it were scalable with powerconsumption.

with gtx1060 if got the problem, that in my first test, it starts with poerconsumption of 62 w and performance of 9,3 ms, but after some seconds it jumps form 62w to 41 w, and performance drops to 16,2 ms. seems to use a different power state. how can I deactivate this?

[PetitDragon]

guys, my gtx1060 arrvied. Started my testings but I have a problem:

Somone knwos how to deactivate the different powerstates of this card?

gtx1070's performance looked stable and it were scalable with powerconsumption.

with gtx1060 if got the problem, that in my first test, it starts with poerconsumption of 62 w and performance of 9,3 ms, but after some seconds it jumps form 62w to 41 w, and performance drops to 16,2 ms. seems to use a different power state. how can I deactivate this?

try to install NVIDIA Inspector and run the following command line:

nvidiaInspector.exe -forcepstate:0,0

[pirlouy]

Why would you want that ?
If power states are well done, you have nothing to do, you won't see a difference on your screen.

[Sunset1982]

Why would you want that ?
If power states are well done, you have nothing to do, you won't see a difference on your screen.

for normal viewing it would be good. But for my comparison it is not, because I can't compare its performance with different settings to the other cards I've tested.



try to install NVIDIA Inspector and run the following command line:

nvidiaInspector.exe -forcepstate:0,0

And what command I have to write to reset it to default after my testings?

[Asmodian]

Care to explain? Is there a major drawback except performance hit?

Is there a major difference with error diffusion from 1080->2160p?

It is only a matter of diminishing returns and increasing performance requirements.

As the pixels get smaller the benefits of ED over ordered dithering decrease and while the number of pixels goes up the performance hit does too. At UHD the first thing I would drop would be error diffusion in favor of ordered dithering.

[PetitDragon]

And what command I have to write to reset it to default after my testings?

"That's the problem, sir. We don't know." :D

[Asmodian]

And what command I have to write to reset it to default after my testings?

Just reboot.

[Knight77]

ok, I will test it. Card will arrive today



Edit:

Tested for you. File was 1920x1080 (with black bars), 23,976, 8 Bit AVC, D3D11 FSE, 10Bit mode:

27,04 ms / 140 w (with automatically detect hard coded black bars feature & crop black bars)
33,45 ms / 170 w (without detecting hard coded black bars)


tested with 1920x1080 (Avatar Mkv, real 16.9 fullscreen without black bars), 23,976, 8 Bit AVC:

34,1 ms / 172 w

Could you please tell me the GPU/engine load and fan speed? My GTX 960 can manage this kind of settings but with high load and, by consequence, a high fan speed... meaning noise.

[ryrynz]

Could you please tell me the GPU/engine load and fan speed? My GTX 960 can manage this kind of settings but with high load and, by consequence, a high fan speed... meaning noise.

1060 is about twice as fast as a 960, if you've got the cash it's a worthwhile upgrade. It may be better to give yourself headroom and a good cooler to avoid noise/heat.
Anyway.. may wish to avoid excessive talk about the 1060 here.

[robl45]

So I got 3d MVC to work great with MPC-HC and madvr using MVC MKV. However the subtitles are flat. Is there anyway to fix that? It appears its only showing one of the two subtitles needed for depth.

[AntonP]

I have 2 problems with 3D playback using LAV + madvr.
Config i7-2600k + gtx980 + win 8.1, latest nightly builds of MPC-HC and LAV.
1. nvidia refresh rate problem - 23.971 instead of 23.976. in 2D mode i fixed it with 24Hz monitor refresh rate (24p instead of 23p in madvr settings) + reclock
in 3D i cant set 24Hz in madvr - 23p mode switches on automaticaly.... how can i set 24p in 3D?
2. artefacts in 3D mode in some scenes (every time in same moment). red car on screenshot is example of it. in powerdvd 16 all is ok, so its not BD3D problem. I think its decoder's problem, but not sure. how to fix this bug?

no ideas?

[chros]

27,04 ms / 140 w (with automatically detect hard coded black bars feature & crop black bars)
33,45 ms / 170 w (without detecting hard coded black bars)

tested with 1920x1080 (Avatar Mkv, real 16.9 fullscreen without black bars), 23,976, 8 Bit AVC:
34,1 ms / 172 w
170 w???!!!! :)
And what command I have to write to reset it to default after my testings?
Doesn't this work?
nvidiaInspector.exe -setPstateLimit:0,0
Anyway.. may wish to avoid excessive talk about the 1060 here.
Agreed, open a new topic somewhere and link it here.

[Sunset1982]

Ok, last post on the gtx1060 / rx480 topic, let's get back to madvr: (you can write me some pm if you got questions)


here is my upgraded comparison:

http://www.file-upload.net/download-11785367/MadVRComparison.xlsx.html


keep in mind, that this is a custom gtx 1060 (palit superjetstream) with higher clocks than standard 1060 vs. standard RX 480. With that in mind, I think they are very close in terms of madvr performance. RX 480 custom designs may get the same madvr performance as the gtx. Only power consumption is a big plus on the gtx.


Some quick notes:

- the Palit Superjetstream is totally silent. Even under heavy load you can not hear the coolers outside of the computer case.

- under low loads the GTX is using different powerstates. Thats good for normal viewing (saves power) but hard to measure for my comaprison. For example, with low madvr settings the gtx starts with 10 ms rendering times, but after some seconds it switches to 28ms, while using only 1/3 power. I think thats a good thing. RX didn't seem to have such a feature.

- I tested without any tools to deactivate the power states of the gtx. Wrote down the performance within the first seconds. I will repeat the test within the next day with deactivated power saving feature and upgrade my excel file, if anything changes. But I think it is already very precisely.

[pose]

Sunset1982, thanks for the test! Glad that in terms of performance RX480 and GTX1060 are almost the same in madVR. Did you decide which card will stay with you already?

[seiyafan]

Your next target is the new Titan X.

I am just kidding, thanks for the good work!

[Sunset1982]

Maybe I stay with the 1060, because I got it here already and no custom RX is available right now.

If there where a good custom RX model for a price up to 250 € I would go for the rx480, just because of the price and the good amd driver support even for older cards. Only thing that must be investigated is the green screen error with madvr when using HEVC decoder and lav's DXVA.

[robl45]

So I got 3d MVC to work great with MPC-HC and madvr using MVC MKV. However the subtitles are flat. Is there anyway to fix that? It appears its only showing one of the two subtitles needed for depth.

I tried it with Kodi and the subtitles are at the proper depth, is this a limitation with MadVR or mpc-hc?

[robl45]

Regarding the 3D, I use stereoscopic player with "Side By Side" 4k 23p Playback, this makes the movie 1080P (1920x2160) and not half vertical res as most SBS converts.

The reason behind this is because MVC subtitles generally are buggy with MKV because there's no subtitle plane tracking in the MKV container, this will make the subtitles 2D, and if the movie is 16:9 the 3D effect will cut through the subtitles, often disrupting the viewing and making it annoying, if not dangerous for the eyes.

The only way to fix this "bug" on MVC MKVs was by putting the MVC video on the right and the 2D on left with subtitles on both sides and playing it side by side, if you don't have a 4k TV this would cut vertical res in half as said, but it's better than buggy playback.
Another thing is that you can't use .srt subtitles with MVC 3D effect as far as i'm aware, you need the original pgs in .m2ts container with the MVC VUI/SEI.

I wonder if you'll add this onto MadVR, as it would be a great feature if you can't fix the MKV 3D MVC subtitles.

Was this subtitle issue ever fixed and maybe I'm doing it wrong?

[strumf666]

Can madvr work in "portable mode"? MPC-HC and BE,... have lav filters built-in and have working portable versions, the only thing missing is madvr as renderer.