madVR - high quality video renderer (GPU assisted)

[aufkrawall]

No deband:
http://abload.de/thumb/debandofftkrlo.png (http://abload.de/image.php?img=debandofftkrlo.png)

madVR high:
http://abload.de/thumb/madvrhighu5rtr.png (http://abload.de/image.php?img=madvrhighu5rtr.png)

madVR high + Shiandow (power: 0.1):
http://abload.de/thumb/madvrshiandowyqrgv.png (http://abload.de/image.php?img=madvrshiandowyqrgv.png)

The left banding in the lower image half almost disappears, but luckily, almost no further detail. :)

I really like this as a test scene since it includes banding that both algorithms don't catch and also unwanted image changes are quickly apparent.
I'd say most detail is already lost with madVR@high, additional Shiandow with low power doesn't hurt much.

[cyberscott]

Ah ok. So probably no vote from you about old low/medium vs Shiandow?




I did some additional comparisons using old low/medium vs Shiandow. I used Power 0.01 to compare old low and Power 0.25 to compare old medium, just to keep it simple. I also used Margin 0.2 for both comparisons. Saiandow settings made the banding less apparent to my eyes, even at the Power 0.01 setting vs old low. I used the OP of Higurashi no Naku Koro ni Kai upscaled from 480p to 3840x2160 and paused at 40 seconds in. Lots of banding on the blue background to do testing on.

[luk008]

I've just discovered that the bug doesn't happen if you set the display to match the fps of the video. In other words, if I set the screen to 23.976 hz, madVR can fill the queues normally in FSE 10 bits.

[dansrfe]

Is there a methodical way to determine whether flush settings should be used and which ones? I have disabled flush on my GTX 670 yet it seems my GT 750M works better with the default flush, flush/sleep, no flush, no flush settings.

[har3inger]

I replaced by AMD 12.200 Beta drivers with the latest 15.4 (Leshcat version).

Great news!! Finally a driver that provides better performance! Especially for lower resolution videos so it most likely better process Jinc upscaling.

Before/After
288p: 16/15ms
720p: 14/13.4ms
1080p: 11/11ms

As far as DirectX 11 presentation, it still doesn't work. It still gives a black screen.

NNEDI3 still doesn't work either. But this time the setting slightly alters rendering times, but not enough to show NNEDI3 usage.

This is with a Radeon HD 7670M on Windows 7 x64

http://s15.postimg.org/6e553zxpz/NNEDI3.jpg (http://postimg.org/image/6e553zxpz/)

Hello fellow mobile ATI user, could you check these things to see if we have the same problem?

Do you have switchable graphics, or is your 7670M on at all times?

Do you have an OpenCl registry key in HKCU\Software\madshi\madvr\? (I don't, and can't get one to appear)

Do error diffusion or nnedi3 actually work? What if you force your iGPU to render it? (ED works only if iGPU is forced. nnedi3 refuses to work at all. I get black screens when it doesn't work)

Under general settings, does DX11 or a separate device for presentation (and restarting to turn them on) cause black screen with audio? (does for me)

If we have the same problem(s), we could maybe look at possible driver or config solutions that might be responsible.

Win 8.1, R9 M270X, Intel 4000HD, CCC 14.12, Intel driver 10.18.10.3304

[James Freeman]

After some training I learned to clearly differentiate between the two debanding algos.

Some findings about Shiandow's algo:
Shiandow's default settings (0.50, 0.00) are somewhere between Mid and High of madVR in terms of debanding efficiency.
Retains general shape better than madVR, hence less "detail" loss.
Smoother with less dithering/grain visible (I see tiny squares [2x2 pixels] with madVR debanding).

The blending between gradients of madVR debanding adds clearly visible grain to the transitions and retains less of the general shape.
On the opposite, in Shiandow's I can't see the grain (silky smooth) and it retains the general shape a lot better.
In this iteration of madVR (0.88.8) Shiandow's is a clear winner compared to Mid and High.


In Low setting it's the opposite (quick [not thorough] test).
I like madVR's Low better than any other setting of Shiandw's (margin on 0.00).
This grain that removed the general shape on Mid and High is important part of Low.
I did not find any setting on Shiandow's that does a better job than Low.
This without playing with Margin which I don't see if it works at all.

[omarank]

As far as “low” and “medium” presets are concerned, the original debanding algorithms seem to be perfect. I clearly prefer them over Shiandow’s debanding.

For “high” preset I am not sure which one is better. However, I found one case of a poorly processed video where the old “high” preset worked perfectly but Shiandow’s algo could not be tuned to remove banding that much effectively. Please see the screenshots in the links below. I would maintain that the old “high” preset should be tuned if required but not be replaced.

Original Image
http://www.mediafire.com/view/nst4vrzmaope170/original.png

Shiandow 0.1, 0.0
http://www.mediafire.com/view/jyauw3xnxz5q8n8/Shiandow%200.1%2C%200.00.png

Shiandow 0.4, 0.01
http://www.mediafire.com/view/xe6apy95mf5bz6a/Shiandow%200.4%2C%200.01.png

Shiandow 0.6, 0.03
http://www.mediafire.com/view/844c85asacq3kew/Shiandow%200.6%2C%200.03.png

madVR old “high” preset
http://www.mediafire.com/view/pdfipkv918aacts/madVR%20old%20high%20preset.png


Having stated my preference, I should also mention that after the new changes being made, Shiandow’s debanding has become really very good. There is nothing to complain, it’s just that in comparison I am preferring the old presets.

[Ava Pug]

I did a few tests on my HTPC with DX11 and I found I couldn't play anything without getting glitches during play playback (gif like).

I'm having issues with the resolution of the video and the TV. For example if I play a 720p23 video and my TV is at 1080p23 the video will glitch out. If I play a 1080p23 video and my TV is at 1080p23 it will play fine.

I found that if I play any video that's not at the native resolution of my TV it will glitch out. The only time it doesn't glitch out is if the video is at native resolution or if madVR doesn't switch display mode.

DX9 works fine, I'm running default madVR settings, Windows 8.1 x64 and Intel HD4400.

[x7007]

That's due to PotPlayer wanting to draw animations and stuff during paused playback. It's madVR's fault, really, but if you turn the PotPlayer madVR specific OSD off, this problem should go away for now. Of course then you'll probably lose the FSE compatible OSD, as well. The high GPU load in paused state with PotPlayer will probably be fixed in a future build, but it has low priority for now.

Do you use subtitles ? do you pause when subtitles are shown ?

It happened to me with potplayer only with subtitles and when the option Subtitles > Word Searching > Condition > Use when playback is paused

That fixed my issue GPU is 0% when video is paused.
Cpu is 3% 4% randomly max 7% 8% or so.

[tyto]

In D3D11 exclusive (10 bits), the render queue and the upload queue stay at 2-4 / 15. Why 2-4? Am I doing something wrong?

Same here
10bits FSE gives me 4-5 rendered frames, with any settings. Instead with D3D9 I can go up and fill my queue.

I tried changing "max presented frame" from nvidia panel but nothing happens.

OS: Windows 7 64bit
MadVR + mpc-hc 64bit
nvidia 650 ti boost, driver version 350.12

It doesnt really affects me much, since my presented queue is almost full, and I dont have any drop frame

[mogli]

I suppose you've unchecked the option "present a frame for every VSync"? This option is enabled by default. Re-enable it, and the problem will probably go away.
Will try.

EDIT: Happens even with option enabled. Will do more testing to find out the exact conditions.

Secondly, in FSE mode D3D11 allows me to skip some VSyncs without having to present a frame, while still being able to present frames in advance. Which means that GPU performance in D3D11 FSE mode should be slightly better compared to D3D9, if you have a display refresh rate which is higher than the source frame rate.
Does this also hold true while having smooth motion activated?

[x7007]

Guys, any quick tip on how to use Potplayer x64 + Madvr x64 ? cause last time I tried it just didn't find the madvr64 file, potplayer x32 + madvr x86 always work great, I just can't figure how to activate the Madvr x64.

And why do you need to delete the Opencl registry every time you change from x86 to x64, at least from what I read. artefacts with NNEDI3, or more issues ?

Thanks

[nijiko]

this should help a lot:
http://forum.doom9.org/showthread.php?t=172128

only the totally high end displays like samsung SUHD can have a true 10 bit panel.

Mine is Philips'...

If I use a wrong bit depth, is there any adverse impact affect?
Such as I use 10bit setting for an 8bit displayer?

[ikarad]

I have some questions,
1) In all madvr 0.88.build, if I use d3d11 path withtout "present a frame for every vsync", I have lots of presentation glitch in FSE (720p,1080p or 4k movies). If I turn on "present a frame for every vsync", no presentation glitch. (win764,gtx780 with the last driver)
a) Is it a bug or not?
b) Does check this option "present a frame for every vsync" harm video quality or affect smoothness or lantency or other?


2) In exclusive mode settings, I have how many frames shall be presented in advance at 3. What means this options?
I try with 16 but I have the same problem in 1)

3) What means cpu queue size and gpu queue size? Is it necessary to modifiy the defaults even if no problem (12 forcpu and 8 for gpu)?

thanks

[nevcairiel]

I have some questions,
1) In all madvr 0.88.build, if I use d3d11 path withtout "present a frame for every vsync", I have lots of presentation glitch in FSE (720p,1080p or 4k movies). If I turn on "present a frame for every vsync", no presentation glitch. (win764,gtx780 with the last driver)
a) Is it a bug or not?
b) Does check this option "present a frame for every vsync" harm video quality or affect smoothness or lantency or other?

The option is explicitly there to avoid this problem, so if you have the glitches, enable that option, and enjoy glitch-free playback.

[nlnl]

Thank you for the new version of your amazing renderer!
Tested FSE mode D3D11, 10 bit/8 bit.
Path is Madvr 88.8, LAV, MPC-BE, Win7, AMD 14.5, Pioneer PDP LX5090.

FSE mode D3D11 does work but using 10 bit output it takes much time to switch to windowed mode (3 seconds of black screen),
BTW Pioneer PDP LX5090 does support 10 bit input (Thank you for the new version of your amazing renderer!
Tested FSE mode D3D11, 10 bit/8 bit.
Path is Madvr 88.8, LAV, MPC-BE, Win7, AMD 14.5, Pioneer PDP LX5090

FSE mode D3D11 does work but using 10 bit output it takes much time to switch to windowed mode (3 seconds of black screen).

BTW Pioneer PDP LX5090 does support 10 bit input (greyscale ramp test done). Old is gold :))
10 bit input is a little bit better then 8 bit + dithering (2x zoom). Self-hypnosis? :)

[x7007]

Can anyone help me with the 10bit and madvr x64 + potplayer ?

I don't know how to check if I have 10bit. what settings should I check in madvr and potplayer and what should I see in the CTRL + J/
I can't run madvr x64 with potplayer x64, it doesn't detect it

[noee]

This link (http://forum.doom9.org/showthread.php?t=172128) by James Freeman has all you need for 8/10bit testing. AFAIK, in potplayer, just make sure you select madVR as the video renderer. Make sure you have run install.bat from the madVR folder.

[huhn]

Mine is Philips'...

If I use a wrong bit depth, is there any adverse impact affect?
Such as I use 10bit setting for an 8bit displayer?

you can get banding.
in best case it will look like 8 bit with dithering.
you have to figure this out by your self i don't have your display.

[kostik]

New "GeForce 352.84 Windows 10TP Driver " allows selecting bit depth finally . I only get 8bpc or 12bpc and my Pioneer plasma changes to 36bit mode (previously had to add registery keys to enable 36bit).
Will test the PQ.

[huhn]

New "GeForce 352.84 Windows 10TP Driver " allows selecting bit depth finally . I only get 8bpc or 12bpc and my Pioneer plasma changes to 36bit mode (previously had to add registery keys to enable 36bit).
Will test the PQ.

352.63 could do this already.

[James Freeman]

Hope this Nvidia driver with the Bit Depth options comes to Windows 7... Probably not.

[huhn]

Hope this Nvidia driver with the Bit Depth options comes to Windows 7... Probably not.

it's not really needed.

if nvidia gets 10 bit input it will switch to 12 bit output. even with way older driver. so nothing really to win if you are right and nvidia isn't dithering.

and i just found out AMD can send 6 bit. i tried it with a display port connection.

[aufkrawall]

New "GeForce 352.84 Windows 10TP Driver " allows selecting bit depth finally . I only get 8bpc or 12bpc and my Pioneer plasma changes to 36bit mode (previously had to add registery keys to enable 36bit).
Will test the PQ.
Borderless and FSE finally works with this WDDM 2.0 driver. :)
No more tearing/jittering for me.

[Kirk Lazarus]

Hi guys,
I use my Htpc (win 7 64bit, Nvidia GTX 970 (latest drivers), MPC HC 64bit and madvr) to see blu ray films to my UHD Tv.
This is my main use.

With the last madVR 0.88.8 I've found a fantastic picture quality with those settings:

PROCESSING:
atifact removal > reduce banding artifacts > settings; low + high

SCALING ALGORITHMS:
chroma upscaling > jinc 3 taps AR
image doubling > luma res. > always NNEDI3 32 neurons
image upscaling > Jinc 3taps AR + scale in linear light
upscaling refinement > Super Res > (default settings) + refine image after every 2x upscaling step + apply superRes first

RENDERING:
dithering > error diffusion option2 > use colored noise + change dither for every frame
trade quality for performance > all unchecked

Unfortunately I've high rendering times that produce too many droped frames:
https://dl.dropboxusercontent.com/u/261382755/logx.jpg

Francly I don't know If my chain is all in 64bit ... I've do not idea how to verify this ...

Anyway there is a work around to maintain those setting and lower the rendering times??

I'm thinking to remove the fan from my gtx 970 and going to liquid cooling system ...
In this case do you think that I can overclock my gpu?
With an overclocked GPU I can lower the rendering times a lot?

I was thinking also to buy a second gtx 970 and going in SLI mode but I was reading that madVR can't take advantage on two GPU in SLI mode ....

any other idea??

[Anime Viewer]

Hi guys,
I use my Htpc (win 7 64bit, Nvidia GTX 970 (latest drivers), MPC HC 64bit and madvr) to see blu ray films to my UHD Tv.
This is my main use.

With the last madVR 0.88.8 I've found a fantastic picture quality with those settings:

RENDERING:
dithering > error diffusion option2 > use colored noise + change dither for every frame
trade quality for performance > all unchecked

Unfortunately I've high rendering times that produce too many droped frames:


Anyway there is a work around to maintain those setting and lower the rendering times??

I'm thinking to remove the fan from my gtx 970 and going to liquid cooling system ...
In this case do you think that I can overclock my gpu?
With an overclocked GPU I can lower the rendering times a lot?

I was thinking also to buy a second gtx 970 and going in SLI mode but I was reading that madVR can't take advantage on two GPU in SLI mode ....

any other idea??

First, I recommend putting a check back in the don't use linear light for dithering check box in trade quality for performance section. It should lower your render times from 40s to 20s without you noticing any visual difference. Second, you probably don't need to be using error diffusion. Watching blu-ray on a high resolution screen you already have a source (the blu-ray) and something with a bit-depth reproduction (the tv) that don't reproduce much of a dither pattern/noise as it is. Try ordered dithering instead. You shouldn't notice any significant difference. There are other unnecessary settings you have enabled, but start with those two.

[cyberscott]

Hi guys,
I use my Htpc (win 7 64bit, Nvidia GTX 970 (latest drivers), MPC HC 64bit and madvr) to see blu ray films to my UHD Tv.
This is my main use.

With the last madVR 0.88.8 I've found a fantastic picture quality with those settings:

PROCESSING:
atifact removal > reduce banding artifacts > settings; low + high

SCALING ALGORITHMS:
chroma upscaling > jinc 3 taps AR
image doubling > luma res. > always NNEDI3 32 neurons
image upscaling > Jinc 3taps AR + scale in linear light
upscaling refinement > Super Res > (default settings) + refine image after every 2x upscaling step + apply superRes first

RENDERING:
dithering > error diffusion option2 > use colored noise + change dither for every frame
trade quality for performance > all unchecked

Unfortunately I've high rendering times that produce too many droped frames:
https://dl.dropboxusercontent.com/u/261382755/logx.jpg

Francly I don't know If my chain is all in 64bit ... I've do not idea how to verify this ...

Anyway there is a work around to maintain those setting and lower the rendering times??

I'm thinking to remove the fan from my gtx 970 and going to liquid cooling system ...
In this case do you think that I can overclock my gpu?
With an overclocked GPU I can lower the rendering times a lot?

I was thinking also to buy a second gtx 970 and going in SLI mode but I was reading that madVR can't take advantage on two GPU in SLI mode ....

any other idea??

I have a GTX 970 in one of my HTPC rigs, also outputting to a UHD TV.
Some of your settings are a little too aggressive and are killing your rendering times.
My main settings for 1080p content upscaled to 4K are as follows..:
SCALING ALGORITHMS:
chroma upscaling > jinc 3 taps AR
image doubling > luma res. > always NNEDI3 32 neurons
image upscaling > Lanczos AR 4 Tap. (linear light should NOT be used when upscaling!)
upscaling refinement > LumaSharpen Default with experimental limiter checked.

RENDERING:
dithering > Ordered Dithering > use colored noise + change dither for every frame
trade quality for performance > all unchecked
..
This should get you to around a 25ms rendering time. Hope this helps!

[ikarad]

Hi guys,
I use my Htpc (win 7 64bit, Nvidia GTX 970 (latest drivers), MPC HC 64bit and madvr) to see blu ray films to my UHD Tv.
This is my main use.

With the last madVR 0.88.8 I've found a fantastic picture quality with those settings:

PROCESSING:
atifact removal > reduce banding artifacts > settings; low + high

SCALING ALGORITHMS:
chroma upscaling > jinc 3 taps AR
image doubling > luma res. > always NNEDI3 32 neurons
image upscaling > Jinc 3taps AR + scale in linear light
upscaling refinement > Super Res > (default settings) + refine image after every 2x upscaling step + apply superRes first

RENDERING:
dithering > error diffusion option2 > use colored noise + change dither for every frame
trade quality for performance > all unchecked

Unfortunately I've high rendering times that produce too many droped frames:
https://dl.dropboxusercontent.com/u/261382755/logx.jpg

Francly I don't know If my chain is all in 64bit ... I've do not idea how to verify this ...

Anyway there is a work around to maintain those setting and lower the rendering times??

I'm thinking to remove the fan from my gtx 970 and going to liquid cooling system ...
In this case do you think that I can overclock my gpu?
With an overclocked GPU I can lower the rendering times a lot?

I was thinking also to buy a second gtx 970 and going in SLI mode but I was reading that madVR can't take advantage on two GPU in SLI mode ....

any other idea??
Try to disable image doubling (nnedi32). It's very demanding.

[leeperry]

Do you have the Windows 7 Platform Update installed? That's a requirement for madVR's D3D11 presentation path.
Touché, Coulé...I saw you recommending it but this wasn't clear to me to be a prerequisite to D3D11 in mVR :o

For some reason this update refuses to install claiming that it wasn't meant for my OS huh, anyway I don't like Aero for all the aforementioned reasons and the zombie frames in mVR during FSE/FSW transitions will always be a complete deal breaker.

I'm very impressed by monostatic LL ED2@8bit so no biggy, either I'll slipstream all updates into a new W7 ISO and reinstall or simply upgrade to W10 but PotP would apparently be unusable in 10bit anyway and the W10 upgrade would mostly depend on whether zombie frames would still be there...as they would appear to be related to Aero to begin with :confused:

There wasn't even flashing black frames on XP, that was great.

There are different SR defaults for NNEDI3 compared to NEDI. Have you adjusted them? Also you can tame SR further down, if you still find it too sharp.
I just ran a lot of comparisons on SD and 720p anime & live action movies @1080p and my impressions haven't changed, I've messed with all the SR settings and still find the EE introduced by NNEDI3 unacceptable and artificial looking compared to NEDI, which mostly looks like a sharper J3AR.

I only find NNEDI3 useful on extreme lowres such as 1/4 DVD because NEDI struggles to detect the right edges when NNEDI3 cuts through the noise like butter, finally a good reason for me to create profiles in mVR..also meaning that I'll still depend on the 13.12 drivers duh.

Here's an extreme scenario 40" 720p sample (https://mega.co.nz/#!C5BzAbqJ!od8S-ee7ZxM0v-Y0j3Eg8Xtw1O9QrK_oQXwwH6Vj6lQ) from this movie (http://www.imdb.com/title/tt1043852/), I do own the original BD so I could compare them but with NNEDI3@CR AR LL with the NNEDI3 SR defaults@ED2-8bit, EE is seriously too strong to my eyes as if all background elements in the room were right on the front, it literally kills perspective when OTOH NEDI with the SR non-double defaults looks fantastic to me :cool:

I've also tried the chroma upscaling SR and I don't think I like it, vanilla J3AR all the way.

Do you use subtitles ? do you pause when subtitles are shown ?

It happened to me with potplayer only with subtitles and when the option Subtitles > Word Searching > Condition > Use when playback is paused

That fixed my issue GPU is 0% when video is paused.
Cpu is 3% 4% randomly max 7% 8% or so.
I hardly ever use subs and disabling this option doesn't change anything regarding high GPU usage when pausing mVR on my box.

[MysteryX]

Do you have the Windows 7 Platform Update installed? That's a requirement for madVR's D3D11 presentation path.
Yes that update is installed.

[JarrettH]

chroma upscaling > jinc 3 taps AR


I'd change this to Bicubic75 with AR too. Many people read that Jinc is best for chroma, but it's almost imperceptibly better when comparisons took place.

[XRyche]

madshi, am I correct in assuming you have no intention of keeping both madVR's and Shiandow's debanding algorithms? All this user testing is to determine which you are going to keep. If by some odd chance you happen to keep both algorithms could you link them to madVR's keyboard shortcut keys. I know you probably aren't keeping both but I wanted to let you know the reason for this otherwise seemingly useless question.

[JarrettH]

Edit: testing something before confirming...

[MysteryX]

There appears to be a bug with the profile selection.

When I open this video on 1080p display, it selects "Laptop 288p" instead of "TV 1080p". If I put full-screen mode ON and then back OFF, then it has the right profile "TV 1080p".

http://s14.postimg.org/s4afurnrh/Profile_Selection.jpg (http://postimg.org/image/s4afurnrh/)

[QBhd]

1625-297<1366... so no, there is nothing wrong with madVR.... None of your rules will match, so it picks the first rule.

[nijiko]

you can get banding.
in best case it will look like 8 bit with dithering.
you have to figure this out by your self i don't have your display.

I've tried both 8bit and 10bit settings, but I can't find any diff. from them.
And not find banding as you said.

[huhn]

I've tried both 8bit and 10bit settings, but I can't find any diff. from them.
And not find banding as you said.

than play it save and stick to 8 bit + dithering. but 10 bit should work fine too.

[ADude]

BTW, the reason why I asked about change log for previous builds, was just clearly illustrated. I was busy for a couple of days, come back, and there are 40 new pages, and the version has gone from 0.88.2 to 0.88.8 ! Of course, that is a good thing, but it is helpful to be able to see what changes occurred in the other 5 versions (since I don't have time for 40 pages! :) ).

So, that is why I previously commented that the complete change log can be found at:

http://www.videohelp.com/software/madVR/version-history#changelog

And I think that link should be put into the first post.

Before someone says yet again that the whole changelog is in the download, please note that sometimes you want to see the changelog in order to see whether you want to download.

[6233638]

* added support for native 10bit output:eek: Thank you! I has been a long time coming, but the results speak for themselves. The 10-bit output is noticeably improved over 8-bit on my display.

I understand that FSE mode is required to get a 10-bit output. However, is it necessary to use a DX11 windowed mode?
I still prefer the old windowed mode since it responds immediately to inputs, instead of the delay that you get from presenting several frames in advance.

However, with active 10bit output (in D3D11 FSE mode) error diffusion is disabled automatically and ordered dithering is used instead. The reason for that is that at 10bit, the quality of the dithering should play no role for image quality, anymore. This only applies to 10bit FSE mode, though. In windowed mode (which is always max 8bit) error diffusion is still used.
I would definitely prefer that this was a "trade performance for quality" option instead of being forced.

When I compare 8-bit and 10-bit, there is a clear improvement.
However, with 8-bit using Error Diffusion I was actually seeing that some tests seemed to be smoother in 8-bit rather than 10-bit.
After having caught up on this topic now (a lot has happened in the last two weeks!) it would seem that what I'm probably seeing is the difference between Error Diffusion and Ordered Dither.
So although you may expect there to be no visible differences, in some specific tests - and perhaps only with displays of a certain type, over a certain size, or contrast ratio - it seems as though the type of dither in use may still matter.

Without currently being able to select Error Diffusion, and only having a keyboard shortcut to toggle dither on/off (which is a noticeable difference even in 10-bit) rather than switch between dither algorithms, it is very difficult to be certain that it's the dither algorithm here, and not some issue with the display itself that the higher noise level of a dithered 8-bit signal was masking.

Rather than use error diffusion with 8-bit and ordered dither with 10-bit, what I might actually prefer is to use the profile system to specify different dither algorithms for the 8/10-bit outputs using something like:
if (fsemode) "error diffusion"
else "random dither"

With a 10-bit output it seems as though the smoother error diffusion should be all that is necessary.
If I am viewing something in windowed mode (8-bit) then I am less likely to care about the quality, and the "rougher" random dither may actually be beneficial for masking banding.

I'm already using profiles to specify different types of dither for different content (I can't use error diffusion with 60 FPS videos) so this would only be a minor tweak of my existing settings.

2) Some users enabled chroma doubling because they thought it would benefit image quality, when they should have increased luma doubling neurons instead. So by removing the chroma doubling options I force users to make better decisions.For game footage rather than film - even encoded to 4:2:0 - NNEDI3 chroma doubling can make a noticeable improvement compared to other chroma scaling options. I suspect animated content would also benefit.
So I am happy to see that the option has returned. (even though I missed the builds in which it was removed) :)

But I'm willing to bow to the majority. What does everybody else say? Should I "allow" Jinc3 to be used for chroma doubling? Instead of blocking it and using Catmull-Rom instead?I think that any time you block a setting from working for performance reasons, it should be a "trade quality for performance" option, rather than something which happens silently.

If I have Jinc selected as my scaling algorithm, that's what I expect it to be scaled by.
I agree that it may not be necessary, but I don't like the idea of selecting Jinc scaling, and having madVR apply Catmull-Rom instead.

I have some highly specific presets that I use for certain types of content (that switches automatically due to the excellent profile system) which you might expect to be the "wrong" combination of scaling settings unless you actually saw them paired with that specific type of content, and I'd hate for any of them to break because it was being disabled for performance reasons.

One example might be a preset where NNEDI3 image doubling is used, and then SoftCubic upscaling is used to scale that up to 1080p. You might think that it's a waste of performance to be using NNEDI3 only to blur the result with SoftCubic, but for that specific content it's the best looking result, and the performance hit for using NNEDI3 "unnecessarily" doesn't matter on my system.

Btw, can we get dithering for every frame presented? With Dx11 path and "present every vsync" option that should beat any FRC.That would be a nice option for people with refresh rates higher than the source framerate. (e.g. 144Hz displaying 24p)

2) Or you have the number of video frames which shall be presented in advance set to 16. I've just found out after releasing v0.88.7 that 15 is that max I can do. Using 16 means the D3D11 device creation fails. So lower this option to 14 (15 isn't currently supported), and D3D11 should start working again.Though the 10-bit output was working, I had to reduce this from 16 to 12 on my system to stop the queues being almost empty in the 10-bit DX11 FSE mode.
Only 10-bit was affected, 8-bit DX11 FSE worked just fine at 15. (16)

Windows 8.1 x64, GTX570 (350.12 and 347.88 tested), 32-bit madVR 0.88.8

EDIT: Actually this didn't fix the problem.
After 20-30 seconds of playback, every 5 seconds or so it started switching between full queues and almost-empty queues again, and stuttering while reporting a presentation glitch.
Reducing this even further (8) may have fixed the issue but I'd have to watch something reasonably long to confirm it.

EDIT2: After 30 minutes of playback, the queues remained full and there were no presentation glitches.
That may not solve it on everyone's system, but seems to have fixed the problem for me.


Haven't had a chance to look over any of the new scaling or debanding options at all. Seems like there is a lot to go over!

[huhn]

For game footage rather than film - even encoded to 4:2:0 - NNEDI3 chroma doubling can make a noticeable improvement compared to other chroma scaling options. I suspect animated content would also benefit.
So I am happy to see that the option has returned. (even though I missed the builds in which it was removed) :)

do you have some screen that are not from pixelart 320x200 games?

[nevcairiel]

Though the 10-bit output was working, I had to reduce this from 16 to 12 on my system to stop the queues being almost empty in the 10-bit DX11 FSE mode.
Only 10-bit was affected, 8-bit DX11 FSE worked just fine at 15. (16)

Windows 8.1 x64, GTX570 (350.12 and 347.88 tested), 32-bit madVR 0.88.8

EDIT: Actually this didn't fix the problem.
After 20-30 seconds of playback, every 5 seconds or so it started switching between full queues and almost-empty queues again, and stuttering while reporting a presentation glitch.
Reducing this even further (8) may have fixed the issue but I'd have to watch something reasonably long to confirm it.

EDIT2: After 30 minutes of playback, the queues remained full and there were no presentation glitches.
That may not solve it on everyone's system, but seems to have fixed the problem for me.

Did you monitor GPU memory usage? The 570 didn't have a whole lot of memory if I remember correctly, and 10-bit surfaces need more memory than 8-bit did.

[6233638]

Did you monitor GPU memory usage? The 570 didn't have a whole lot of memory if I remember correctly, and 10-bit surfaces need more memory than 8-bit did.I thought it might be that, but it was only using 929/1280MB.
My testing was only using very basic settings too: 1080p displayed 1:1 with Bicubic 75 chroma, so the render times were around 5ms.
It looks like I can maybe get away with presenting 10 frames in advance, though 12 initially looked like it would be OK too, so I might just leave it at 8 since that's working.
Finished a 2 hour movie with it set to 8 and the buffer was always full, with no presentation glitches.

do you have some screen that are not from pixelart 320x200 games?I don't off-hand, but can probably get some tomorrow.
It was mainly 720p footage of more modern games that I was testing that with.
NNEDI3 really seems to clean up icons using fine lines or colored text.

[webs0r]

Hello, 2 things I wanted to check from this 0.88 version -

(1) Visual artifacts (little blocks) appearing in BT.601 vids
I found that with the new madVR v0.88 there are visual glitches appearing - like little black squares in videos. It seems to be only BT.601/SMPTE C videos. They disappear and re-appear in different places.

I first noticed it while playing the smallramp.ytp from madtestpatternsource. Also occurs while the video is paused.
It happens regardless of the madvr calibration setting (3dlut or disable).
madVR d3d9 windowed mode (8 bit), all settings set to don't flush (doesn't seem to matter), exclusive/windowed mode doesn't matter, still happens.

(2) MadTPG & Argyll - how do I get madTPG to default to the largest "Image Area"?
When I run dispcal and/or dispread the image area keeps resetting to the smallest size possible, even after I drag the slider. Had a quick check, can't seem to find how to alter this behaviour?

System details:
Win7 x64
nvidia GTX680, driver 350.12. Also occurred on v347.
mpc-hc 1.7.8.162 beta
Argyll 1.7.0

All advice appreciated!!

[Anime Viewer]

When I compare 8-bit and 10-bit, there is a clear improvement.
However, with 8-bit using Error Diffusion I was actually seeing that some tests seemed to be smoother in 8-bit rather than 10-bit.
After having caught up on this topic now (a lot has happened in the last two weeks!) it would seem that what I'm probably seeing is the difference between Error Diffusion and Ordered Dither.
So although you may expect there to be no visible differences, in some specific tests - and perhaps only with displays of a certain type, over a certain size, or contrast ratio - it seems as though the type of dither in use may still matter.

What are you basing your opinion on (your eyes), or something else?
What is the display you are using to make the that conclusion? (Screen size, DPI, resolution, etc).
I would vote against it being in the trade quality for performance screen (which is already packed) unless others can validate your findings. Too many people seem to jump into that screen, and not knowing any better unchecking everything in there thinking it will give them a better picture (and that there system can handle it), and that is rarely ever true. Then they come here complaining about dropped frames and high render stats (among other things).

[huhn]

Hello, 2 things I wanted to check from this 0.88 version -

(1) Visual artifacts (little blocks) appearing in BT.601 vids
I found that with the new madVR v0.88 there are visual glitches appearing - like little black squares in videos. It seems to be only BT.601/SMPTE C videos. They disappear and re-appear in different places.

I first noticed it while playing the smallramp.ytp from madtestpatternsource. Also occurs while the video is paused.
It happens regardless of the madvr calibration setting (3dlut or disable).
madVR d3d9 windowed mode (8 bit), all settings set to don't flush (doesn't seem to matter), exclusive/windowed mode doesn't matter, still happens.

(2) MadTPG & Argyll - how do I get madTPG to default to the largest "Image Area"?
When I run dispcal and/or dispread the image area keeps resetting to the smallest size possible, even after I drag the slider. Had a quick check, can't seem to find how to alter this behaviour?

System details:
Win7 x64
nvidia GTX680, driver 350.12. Also occurred on v347.
mpc-hc 1.7.8.162 beta
Argyll 1.7.0

All advice appreciated!!

there is a bug in the current nvidia driver that produce issue if you use nnedi3. sound like you have encounter this bug.

you can work around that bug by installing new nvidia driver and using madVR x64 before x86 sounds strange but this works.

[6233638]

What are you basing your opinion on (your eyes), or something else?
What is the display you are using to make the that conclusion? (Screen size, DPI, resolution, etc).
I would vote against it being in the trade quality for performance screen (which is already packed) unless others can validate your findings. Too many people seem to jump into that screen, and not knowing any better unchecking everything in there thinking it will give them a better picture (and that there system can handle it), and that is rarely ever true. Then they come here complaining about dropped frames and high render stats (among other things).As I said: without the option to use error diffusion with the 10-bit output, or a shortcut to change the type of dither being used without exiting FSE, it's very difficult to be entirely certain.

But while 10-bit is largely a big improvement over 8-bit on my TV (undithered, there are clearly four times as many steps) there were some tests that I set up which appeared to be slightly better with 8-bit.
Since 10-bit generally looked best in the majority of my tests, I suspect that the difference in dither being used (ED when outputting 8-bit, OD when outputting 10-bit) could be the cause.

Of course it may not be. But if I select Error Diffusion, I would like madVR to use Error Diffusion.
It is easy enough to select another algorithm, or to create a profile which switches to something else when entering FSE mode.

But for me, it is FSE mode (10-bit) where I would be most likely to use Error Diffusion.
With an 8-bit output, Error Diffusion can actually be too smooth, and some banding may remain simply due to the limited number of steps.
Random Dither does a better job hiding banding at the cost of some additional noise with an 8-bit output.
With a 10-bit output that should not be necessary.

If performance is bad, I would say that the user needs to change their settings, not that the settings should lie about what they do.
Perhaps the default should change, or there could be presets that users could select, but I don't like the options saying that one thing is selected, while doing something else.
Or removing options which have a high performance cost because someone may actually use it.

[XMonarchY]

Is there any reason to still include madLevels Tweaker? nVidia now has a fully functional setting that provides full range 0-255 over HDMI in videos and games. Is there any reason to use it?

Another question I had - does MPC-HC override madVR refresh rate settings? For example, my TV supports 23hz and 24hz modes, but they are not exactly 23.976hz because of my videocard. If I create a custom 23.976hz mode in MPC-HC x64 1.7.8 build 191, then will that 23.976hz apply or will madVR just use my TV's default 23hz mode? I also tried to make a 23.976hz custom resolution in nVidia CP, but it creates a 24.000hz resolution instead. How can I force it to use 23.976hz exactly?

[nijiko]

than play it save and stick to 8 bit + dithering. but 10 bit should work fine too.

Oh, I see. Thank you!

[Scyna]

Not sure if its just me but when I played something on my tv which is 1080p thru hdmi I can't fill render queues. When I use my 1080p monitor thru dvi it works fine. This bug only happens with dx11 renderer. This is on amd 290.

[webs0r]

there is a bug in the current nvidia driver that produce issue if you use nnedi3. sound like you have encounter this bug.

you can work around that bug by installing new nvidia driver and using madVR x64 before x86 sounds strange but this works.

Ah thanks huhn... This worked a treat.

Also I figured this out:
(2) MadTPG & Argyll - how do I get madTPG to default to the largest "Image Area"?
When I run dispcal and/or dispread the image area keeps resetting to the smallest size possible, even after I drag the slider. Had a quick check, can't seem to find how to alter this behaviour?

I previously had tried the -P switch (for dispcal) but didn't use big enough values and didn't notice the change!
Use -P0.5,0.5,10,10 (10 means 100%, 1 means 10%). This fills the image area fully just in case anyone was wondering.