madVR - high quality video renderer (GPU assisted)

[SyrupBuccaneer]

A log doesn't really help much if a freeze occurs. What would help is a freeze report which you can usually create by pressing Ctrl+Alt+Shift+Pause/Break. After you press those keys, a freeze report file should be created on your desktop. However, for the freeze report to be useful, PotPlayer would have to contain (or be distributed with external) debug information. Not sure if that is the case.

Took a while to show up again but it did. Attached

[Werewolfy]

Compare these, you'll see maxdif at 2.1 is slightly softer. It's not a negative side effect, just a preference for a less soft appearance.

2.1
http://imageshack.us/a/img189/4088/878x.th.png (http://imageshack.us/photo/my-images/189/878x.png/)

1.9
http://imageshack.us/a/img571/5586/e5rp.th.png (http://imageshack.us/photo/my-images/571/e5rp.png/)

I had to look very very hard to see a difference but yes 2.1 is slighlty softer than 1.9. But we can't say it blurs details and there's also slighlty more banding with 1.9.

Opinions of the other testers? Personnaly, I would choose 2.0 for maxDif, I think 1.9 is just a bit too low. I don't mind either if the final choice is 2.1.

[ryrynz]

I had to look very very hard to see a difference but yes 2.1 is slightly softer than 1.9. But we can't say it blurs details and there's also slighlty more banding with 1.9.


When you're changing one setting by only this much it's subtle. I might like a 1.8 setting, but I didn't want to push the favored preset too far.
I just wanted something to reduce this overall softening I was seeing and this does that without derailing the settings too much.

Also just FYI, this image was resized 3x and the image had it's brightness boosted to show more clearly what was happening.
The effect you see here is thus amplified, check it out on your favorite testing vid. I don't think debanding strength should be favored here at the low setting.

[THX-UltraII]

ok i try it.

i said output YCbCr is not the ideal way because you asked about a way to output this because you heard you ~beamer?~ got problems with rgb right?

so i talked about about a way to output YCbCr with madvr because i heard it's possible (i still don't know it's true).

so madvr outputs 8 bit 0-255 rgb at default settings. if your gpu outputs rgb unlimited noting should happen.

so i don't know what your plan is right now.

are you going with your new amd card or not.

than there was something about rgb YCbCr output in intel.

you can just switch to both in the driver (never heard that this is not working)

but outputing rgb 0-255 was a or is a problem with intel there is no option for this (something about "it"or normal content but this does nothing...) and it's a well known problem.

don't mix that up with video output 16-235 or 0-255.

even if you are going to use limited rgb you still have to use a 16-235 to 0-255 (madvr does this y default) because the desktop is 0-255 and both together needs to be limited again. not the best way to do this but yeah there are plenty of people in this forum with much better english and much better knowledge of this topic then me.

so what's your problem in the detail?
Thanks for your reply. I ve decided I m going to let my HTPC do the 2K => 4K upscaling. So the Intel card is not option at all anymore. I returned my HD7950 because it was a dead card. I now got a brand new Asus HD6970 Direct CUII which I OC-ed to 1103Mhz. I hope this card can handle 2K = 4K upscaling. Madshi told me I have to output quadHD on my card and 'zoom in mpc-be'. I can understand the first part (just choose 3840×2160@23 in the video card options) but what is 'zoom in mpc-be'?

[ryrynz]

Another comparing high maxdif values, this time on clean anime content, nothing here I'd really consider banding that needs to be improved, so it's a test to see what's removed.
A maxdif of 1.9 looks a fair bit more untouched, I'd like to enable low by default for all content without it smoothing things, I can see the difference between a maxdif of 1.9 and 2.1 on my 65" plasma.

1.9
http://s6.postimg.org/3rtcrlrnh/1_9.png (http://postimg.org/image/3rtcrlrnh/)

2.1
http://s6.postimg.org/4rjnnb6t9/2_1.png (http://postimg.org/image/4rjnnb6t9/)

None
http://s6.postimg.org/3qjey6ptp/none.png (http://postimg.org/image/3qjey6ptp/)

[Werewolfy]

I can't see a difference even on my tv screen between deband off and maxDif on 2.1... We'll see what the other testers think about that. Anyway, I don't really mind if 1.9 is the final choice.

[THX-UltraII]

Thanks for your reply. I ve decided I m going to let my HTPC do the 2K => 4K upscaling. So the Intel card is not option at all anymore. I returned my HD7950 because it was a dead card. I now got a brand new Asus HD6970 Direct CUII which I OC-ed to 1103Mhz. I hope this card can handle 2K = 4K upscaling. Madshi told me I have to output quadHD on my card and 'zoom in mpc-be'. I can understand the first part (just choose 3840×2160@23 in the video card options) but what is 'zoom in mpc-be'?I just found out that the HD6950 does not support 4K output resolutions. So for now I decided to let the upscaling be done by my SOny VPL-VW1000ES 4K projector.

@Madshi, I asked this before but is there really no way to let madVR not do any upscaling? Because I now will play 1080p content, madVR will do its algo scaling thing and my SOny 4K projector will again do scaling from 2K to 4K

[huhn]

i found a source which is hard banded on purpose:

avgDif maxDif grad Penalty LC Boost
average: 0.6 1.9 1.4 1.2 off 1.9

http://abload.de/thumb/shinlowepcz0.png (http://abload.de/image.php?img=shinlowepcz0.png)
http://abload.de/thumb/shinoffjri98.png (http://abload.de/image.php?img=shinoffjri98.png)

it's 32 mbit avg slow motion and some scene are not banded only the blue ones.

i don'f find a way to set up the max angle so this can be the problem too.

so i think it should be disabled on default.

@THX-UltraII

try a custom 4k res

[mikser]

Hi guys,

I have been following this discussion with great interest, but haven't participated because of lack of opinions (that would add anything substantial to the discussion) on the topic of low settings. But now that you are starting to discuss the higher settings, I feel obligated to intervene as I strongly disagree with Werewolfy suggestion of:

avgDif maxDif grad Penalty LC Boost maxAngle
medium: 1.0 2.8 1.8 1.2 off 1.9 0.18
high: 1.8 3.9 2.4 1.2 off 1.9 0.22

I don't know about other people, but with the content I tested, these settings do very little. Actually they hardly do anything. And pretty much the only thing that works for bad content is increasing the boost and doing it significantly. How much? At the very least it should be 2.5. I have seen cases where there are bonuses for using a value of 4.0. I am sure we could find cases where even 5.5 would improve the picture.

I won't participate in the discussion about the actual numbers as I feel too incompetent and you guys are doing a good job without me anyways. I just wanted to give you my input. You are smart guys, you will find what to do with all this information. :)

Here are a couple of files that you might find useful for testing:
- File1. This file is a decent quality (considering), but has very big problems where correct debanding would really help (74MB): http://www.sendspace.com/file/acmi3y
- File2. Video of really shitty quality (4MB): http://www.sendspace.com/file/6m69ri

There is also a question of "what kind of content/quality should different settings be targeting". If file1 is "medium" and file2 is "high", then there is a huge gap both between them and between "medium" and "low". So maybe we should add at least one more setting, so file1 would correspond to "high" and file2 to "very high"? Or maybe even two more options, with you deciding on low-medium-high settings (for high-quality content you seem to be testing on) and then adding "very high" and "ultra high" for my type of low-quality content?

Do you need some more video content to test with? I can bring some.

[Werewolfy]

Well, I watched your videos, there is a lot of mosquito noise and 8x8 MPEG2 blocks but not so much banding.

It's true that debanding can sometimes reduce those artefacts but it's not its main purpose. For that, madshi has to build a specific tool like deringing and deblocking.

For the medium preset, I tried to keep as much details as I could and to offer a real good debanding with medium quality content. If I increase the values too much, it blurs a lot of details. I could increase the angleBoost but then I had to decrease the other values. angleBoost doesn't affect the whole picture so I think it's a better idea to keep angleBoost around 1.9 and increase the other values so a good debanding is on even when angleBoost doesn't work.

For the high preset, I don't have a lot of very bad files. It could be higher and I think there is a good chance it will be.

[Thunderbolt8]

does anyone know what nvidias G-Sync function is for? because with the new drivers, G-SYNC is the new default option for global settings and this does also apply for mpc-hc unless this is being changed manually to application controlled now.

so is it better to set vertical sync to application controlled with the new drivers`?

[6233638]

I'm not sure that I'll be able to contribute to med/high settings until the weekend again, as things are quite busy for me right now.

One thing I have noticed though, is that debanding happens before scaling. This is normally fine, but can look quite bad with low resolution content.

I wonder if it would be possible to deband after scaling below a certain resolution? (or possibly at all resolutions?) I realise that this may cause some problems with the debanding algorithm though. (as you no longer have hard edges on it)

Opinions of the other testers? Personnaly, I would choose 2.0 for maxDif, I think 1.9 is just a bit too low. I don't mind either if the final choice is 2.1.I can't tell if that's just a really low resolution source scaled up, or if it was scaled up after the fact for clarity, but I barely see anything different between them. Difficult to say without the video.

Another comparing high maxdif values, this time on clean anime content, nothing here I'd really consider banding that needs to be improved, so it's a test to see what's removed.
A maxdif of 1.9 looks a fair bit more untouched, I'd like to enable low by default for all content without it smoothing things, I can see the difference between a maxdif of 1.9 and 2.1 on my 65" plasma.In these examples, where exactly are you seeing differences?

What display is it you have, and are you doing any kind of post-processing? (sharpening etc?) I'm barely seeing any difference between your images.
Plasmas are not the best thing to test with, as they are not capable of true 8-bit gradation, so you might be seeing contouring or other effects which are not in the actual image.

I do however see blocking in the center of the image which debanding has taken care of.

i found a source which is hard banded on purpose
...
so i think it should be disabled on default.Considering that most sources are not like this, I'm not sure I agree.

The default setting should work well with most sources; which the current settings seem to do.

There is also a question of "what kind of content/quality should different settings be targeting".I have a few things in mind, but for "high" I have been targeting YouTube HD (typically 720p) quality content as that seems to be a common source, and is quite highly compressed.

[turbojet]

I've had good luck with 2.0/4.0/2.0 without much detail loss on a still picture and no noticeable loss during playback. Gradient seems to be the main factor of detail loss but it's helpful at conservative values at the same time. NonGradientPenalty has little effect but it's not always positive so decided to ignore it. LocalContrast doesn't show any noticeable difference during playback and not worth the almost doubled gpu load. AngleBoost has an effect but sometimes leads to detail loss and not worth the more than triple gpu load.

I haven't done thorough testing with troublesome samples, trying to keep it simple, watching mainly HDTV and occasional SD on a 22" 1680x1050 monitor and some SD on a 32" CRT. Probably atypical but it's all I'll have for the foreseeable future.

[kasper93]

avgDif maxDif grad Penalty LC Boost maxAngle
medium: 1.0 2.8 1.8 1.2 off 1.9 0.18
high: 1.8 3.9 2.4 1.2 off 1.9 0.22

For high I'd bump gradient, even something extreme like 3.9-4.2... We have LOW which is very week and preserve almost all details. So high should be exact opposite I mean remove all banding no matter of details ;p

And generally I think we should try to find values for high preset (ie. remove banding at all cost) and make medium with average values between low and high.

So my proposal is that.
avgDif maxDif grad Penalty LC Boost maxAngle
medium: 1.2 2.8 2.7 1.2 off 1.9 0.16
high: 1.8 3.6 3.9 1.2 off 1.9 0.22

not sure about Boost, though... Please note that I have limited range of samples with heavy banding and my results may be inaccurate

@madshi

Maybe you could store presets in register? So everyone could adjust them.

[ryrynz]

In these examples, where exactly are you seeing differences?


Just about everywhere really. Here is a comparison of maxdif 1.9 to 2.1 using Beyond Compare's tolerance mode which indicate matches, unimportant differences, and important differences.
I've disabled dithering because it changes just way too many pixels to do proper comparisons.

http://s6.postimg.org/barnrl1z1/Image8.jpg (http://postimg.org/image/barnrl1z1/) (link will take you to a large thumbnail, click again to view the full PNG)

Black signifies unchanged pixels, blue is unimportant (but still changed), red is more important (but still reasonably minor considering the threshold I set)
As you can see there is quite a significant change, 40.76% of the entire image is affected in some way just by changing maxdif down from 2.1 to 1.9.


What display is it you have, and are you doing any kind of post-processing? (sharpening etc?) I'm barely seeing any difference between your images.


A Panasonic VT50 and yes I am using image sharpening in this particular image but not with the previous one I posted, regardless of whether I have Avisynth post processing enabled I still see the same differences.


Plasmas are not the best thing to test with, as they are not capable of true 8-bit gradation, so you might be seeing contouring or other effects which are not in the actual image.


I see the same changes on my Dell U2412 too, they are subtle but they're there.


I do however see blocking in the center of the image which debanding has taken care of.


Yup and it's taken care of nicely with both settings. I don't mind if banded areas are smoother, but when what I consider to be non banded areas are changed, I really would like to see that minimized as much as possible, minor or not.

Here's another tolerance mode comparison of the same screenshot I used above.

The original frame, no debanding.
http://s6.postimg.org/97h8jx265/none.jpg (http://postimg.org/image/97h8jx265/)

Low preset with maxdif 2.1
http://s6.postimg.org/h1mfj1zct/image.jpg (http://postimg.org/image/h1mfj1zct/)

Low preset with maxdif 1.9
http://s6.postimg.org/5dsdoia7x/image.jpg (http://postimg.org/image/5dsdoia7x/)

Tolerance mode check no deband vs low with maxdif of 2.1
http://s6.postimg.org/x2f0w0x8d/none_to_21.jpg (http://postimg.org/image/x2f0w0x8d/)

Tolerance mode check no deband vs low with maxdif of 1.9
http://s6.postimg.org/we66d2yil/none_to_19.jpg (http://postimg.org/image/we66d2yil/)

As you can see a maxdif of 2.1 encroaches more on non banded areas than with a maxdif of 1.9

Some numbers for comparison.

no deband vs low preset with maxdif 2.1

1510093 same pixel(s)
787040 unimportant difference pixel(s)
6867 important difference pixel(s)


no deband vs low preset with maxdif 1.9

1522985 same pixel(s)
774876 unimportant difference pixel(s)
6139 important difference pixel(s)

If it didn't affect debanding quality noticeably I'd push for a further maxdif reduction, but I suspect it would all start to add up.

[vivan]

ryrynz,
disabling dithering is wrong. If you doing it for some reason, then what you're doing is wrong too :P
All pixel values are not integers anyway, so their slight change can change their truncated value. If you really, really want to get and compare real values - then you should take several (e.g. 16 for 12 bit precision) screenshots with different dithering patterns and then aproximate image (using higher bitdepth).
Also if those changes are less significant than dithering, then they're not significant at all.

Amount of different pixels doesn't really mean anything. If you want to calculate amount of pixels that were changed by mistake, then you should invent an algorithm that can distinguish detail from banding. Wait... this is is what debanding algorithm is actually trying to do :)
I mean that statement "there's no banding on this image" is subjective, so you can't make any conclusions using objective methods on it. You shouldn't use it as base.
Let's say we have uncompressed source of this video. Using such method you'll make a conclusing that this source is worse than your video, since it's different too.

P.s. I've recently found video that is harmed by such "low" settings, sample (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/Rurouni_Kenshin_-_01.mkv).
No debanding (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_s.png), 1.9 (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_19.png), 2.1 (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_21.png). Even there I don't really see any difference between 1.9 and 2.1.

[ryrynz]

ryrynz,
disabling dithering is wrong. If you doing it for some reason, then what you're doing is wrong too :P


How so? If I enable it it represents changes in the picture and it's random. Disabling it avoids that for a true comparison.

ryrynz,
Also if those changes are less significant than dithering, then they're not significant at all.


Maybe so, but there's a difference I can see time and again, regardless of what tools are used to represent it, I can still see it.

ryrynz,
Amount of different pixels doesn't really mean anything. If you want to calculate amount of pixels that were changed by mistake, then you should invent an algorithm that can distinguish detail from banding. Wait... this is is what debanding algorithm is actually trying to do :)


You've taken one out of three lines I pasted and made it sound like I'm using one of them by itself to make a claim. Wouldn't it make sense that less changes to an image means that it's likely that the image is truer to the original? Especially if my eyes are backing it up?

ryrynz,
I mean that statement "there's no banding on this image" is subjective, so you can't make any conclusions using objective methods on it. You shouldn't use it as base.
Let's say we have uncompressed source of this video. Using such method you'll make a conclusing that this source is worse than your video, since it's different too.


I never said there is "no banding on this image" I understand what can be considered banding, the trick is to work on the settings so that only noticeable or extreme banding is improved upon.. why does it sound like you're fighting against retaining image quality?

ryrynz,
P.s. I've recently found video that is harmed by such "low" settings, sample (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/Rurouni_Kenshin_-_01.mkv).
No debanding (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_s.png), 1.9 (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_19.png), 2.1 (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_21.png). Even there I don't really see any difference between 1.9 and 2.1.

I'll have a look.

[Werewolfy]

Just about everywhere really. Here is a comparison of maxdif 1.9 to 2.1 using Beyond Compare's tolerance mode which indicate matches, unimportant differences, and important differences.
I've disabled dithering because it changes just way too many pixels to do proper comparisons.

http://s6.postimg.org/barnrl1z1/Image8.jpg (http://postimg.org/image/barnrl1z1/) (link will take you to a large thumbnail, click again to view the full PNG)

Black signifies unchanged pixels, blue is unimportant (but still changed), red is more important (but still reasonably minor considering the threshold I set)
As you can see there is quite a significant change, 40.76% of the entire image is affected in some way just by changing maxdif down from 2.1 to 1.9.



A Panasonic VT50 and yes I am using image sharpening in this particular image but not with the previous one I posted, regardless of whether I have Avisynth post processing enabled I still see the same differences.



I see the same changes on my Dell U2412 too, they are subtle but they're there.



Yup and it's taken care of nicely with both settings. I don't mind if banded areas are smoother, but when what I consider to be non banded areas are changed, I really would like to see that minimized as much as possible, minor or not.

Here's another tolerance mode comparison of the same screenshot I used above.

The original frame, no debanding.
http://s6.postimg.org/97h8jx265/none.jpg (http://postimg.org/image/97h8jx265/)

Low preset with maxdif 2.1
http://s6.postimg.org/h1mfj1zct/image.jpg (http://postimg.org/image/h1mfj1zct/)

Low preset with maxdif 1.9
http://s6.postimg.org/5dsdoia7x/image.jpg (http://postimg.org/image/5dsdoia7x/)

Tolerance mode check no deband vs low with maxdif of 2.1
http://s6.postimg.org/x2f0w0x8d/none_to_21.jpg (http://postimg.org/image/x2f0w0x8d/)

Tolerance mode check no deband vs low with maxdif of 1.9
http://s6.postimg.org/we66d2yil/none_to_19.jpg (http://postimg.org/image/we66d2yil/)

As you can see a maxdif of 2.1 encroaches more on non banded areas than with a maxdif of 1.9

Some numbers for comparison.

no deband vs low preset with maxdif 2.1

1510093 same pixel(s)
787040 unimportant difference pixel(s)
6867 important difference pixel(s)


no deband vs low preset with maxdif 1.9

1522985 same pixel(s)
774876 unimportant difference pixel(s)
6139 important difference pixel(s)

If it didn't affect debanding quality noticeably I'd push for a further maxdif reduction, but I suspect it would all start to add up.

I still can't see a difference between your pictures, sorry... I have also tested with a pattern to see very easily if the sharpness of the picture is modified. I have seen no difference between maxDif off and MaxDif on 40.0.

avgDif maxDif grad Penalty LC Boost maxAngle
medium: 1.0 2.8 1.8 1.2 off 1.9 0.18
high: 1.8 3.9 2.4 1.2 off 1.9 0.22

For high I'd bump gradient, even something extreme like 3.9-4.2... We have LOW which is very week and preserve almost all details. So high should be exact opposite I mean remove all banding no matter of details ;p

And generally I think we should try to find values for high preset (ie. remove banding at all cost) and make medium with average values between low and high.

So my proposal is that.
avgDif maxDif grad Penalty LC Boost maxAngle
medium: 1.2 2.8 2.7 1.2 off 1.9 0.16
high: 1.8 3.6 3.9 1.2 off 1.9 0.22

not sure about Boost, though... Please note that I have limited range of samples with heavy banding and my results may be inaccurate


I was maybe too carefull with the high preset but I don't think that the medium preset should be the average between low and high but a solution that eradicates most banding in most sources and preserves most details too.

I have a Blu-Ray Player Sony S790 that has debanding. The low preset is very light, it preserves almost most details (but the debanding is weaker than our low preset :) ). The medium preset is just a bit stronger but the high preset is a lot stronger. It gets rid of almost all banding but the details are... destroyed :D

I modified a little bit my presets :

avgDif maxDif grad Penalty LC Boost maxAngle
medium: 1.1 2.8 1.8 1.2 off 1.9 0.18
high: 1.8 4.0 2.8 1.2 off 1.9 0.25

[ryrynz]

ryrynz,
disabling dithering is wrong. If you doing it for some reason, then what you're doing is wrong too :P


How so? If I enable it, it represents changes in the picture and it's random. Disabling it avoids that for a true comparison.

ryrynz,
Also if those changes are less significant than dithering, then they're not significant at all.


Maybe so, but there's a difference I can see time and again, regardless of what tools are used to represent it, I can still see it.

ryrynz,
Amount of different pixels doesn't really mean anything. If you want to calculate amount of pixels that were changed by mistake, then you should invent an algorithm that can distinguish detail from banding. Wait... this is is what debanding algorithm is actually trying to do :)


You've taken one out of three lines I pasted and made it sound like I'm using one of them by itself to make a claim. Wouldn't it make sense that less changes to an image means that it's likely that the image is truer to the original? Especially if my eyes are backing it up?

ryrynz,
I mean that statement "there's no banding on this image" is subjective, so you can't make any conclusions using objective methods on it. You shouldn't use it as base.
Let's say we have uncompressed source of this video. Using such method you'll make a conclusing that this source is worse than your video, since it's different too.


I never said there is "no banding on this image" I understand what can be considered banding, the trick is to work on the settings so that only noticeable or extreme banding is improved upon.. why does it sound like you're fighting against retaining image quality?

ryrynz,
P.s. I've recently found video that is harmed by such "low" settings, sample (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/Rurouni_Kenshin_-_01.mkv).
No debanding (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_s.png), 1.9 (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_19.png), 2.1 (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_21.png). Even there I don't really see any difference between 1.9 and 2.1.

I'll have a look.

That image has so little detail it's not funny. It doesn't harm it, I can see quite clearly that at maxdif 1.9 it's just more like the original frame, it's quite evident on the skin, at 2.1 the image is blended more in the debanding area
which you prefer because of how bad the gradients are. Honestly, you're just wanting a smoother image for that particular scene and you should be using medium settings for that not low, you shouldn't expect a deband tool to fix Kenshin's splotchy face, that's really the work for a smoother.

Actually misunderstood what you were getting at, will reply further down.

[Ver Greeneyes]

madshi, have you considered implementing ordered dithering for use cases like this, where bit perfect comparisons are desirable? Either that or using a fixed initial seed for the random dithering.

[Werewolfy]

P.s. I've recently found video that is harmed by such "low" settings, sample (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/Rurouni_Kenshin_-_01.mkv).
No debanding (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_s.png), 1.9 (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_19.png), 2.1 (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_21.png). Even there I don't really see any difference between 1.9 and 2.1.

Yes it's a bit too soft on the edges of the hair but avoiding that with debanding is extremely hard. To avoid that, we shoud use these settings :

avgDif maxDif grad Penalty LC Boost maxAngle
0.5 2.1 1.1 1.2 off 1.9 0.10

But with these settings, the debanding is very weak (but not invisible). It might suit for a very low preset, maybe it's not so useless after all.

[ryrynz]

Yeah, but everyone decided that this was the most effective debanding strength for the low preset, personally I think what you gain from debanding being enabled in that video is more than what you
lose from the softness in certain areas. That's the thing generally with image enhancement you win some you lose some, I still think Madshi is still only keen for three presets.

[vivan]

madshi, have you considered implementing ordered dithering for use cases like this, where bit perfect comparisons are desirable? Either that or using a fixed initial seed for the random dithering.IIRC when you save image using Alt+I (Save Image function) madVR dithers on CPU and uses some kind of non-random dithering. But debanding on CPU (or copyback, I dunno) is not implemented (yet?).

On the other hand, even if dithering pattern is not random - even slightest change could change everything.
Imagine that you have values 10.4, 10.5, 10.5, 10.5, 10.5. They would be dithered to 10, 11, 10, 11, 10.
Now you have 10.6, 10.5, 10.5, 10.5, 10.5. They would be dithered to 11, 10, 11, 10, 11 - completely different result.

[ryrynz]

Actually had no idea ALT-I avoided the dithering issue, just done a quick comparison and the images are identical.

[Werewolfy]

Yeah, but everyone decided that this was the most effective debanding strength for the low preset, personally I think what you gain from debanding being enabled in that video is more than what you
lose from the softness in certain areas. That's the thing generally with image enhancement you win some you lose some, I still think Madshi is still only keen for three presets.

I agree, I don't want those settings for a low preset, the debanding strenght is too weak. Actually, madshi thought about a very low preset especially for 4K content.

[Ver Greeneyes]

Imagine that you have values 10.4, 10.5, 10.5, 10.5, 10.5. They would be dithered to 10, 11, 10, 11, 10.
Now you have 10.6, 10.5, 10.5, 10.5, 10.5. They would be dithered to 11, 10, 11, 10, 11 - completely different result.
Seems like that would be implementation dependent.. it's an interesting point though. I was only thinking about dithering after resizing, but I guess that's not the only kind you have to consider here.

[ryrynz]

Actually using Ctrl-I to save the image completely avoids debanding altogether. I'm getting bit identical output regardless of setting.

[madshi]

Took a while to show up again but it did. Attached
I'm sorry but PotPlayer doesn't seem to come with any helpful debug information... :( I'd need a freeze report from a PotPlayer build which as proper debug infos available. I don't know how to get that, though. You may have to contact the PotPlayer developer.

Another comparing high maxdif values, this time on clean anime content, nothing here I'd really consider banding that needs to be improved, so it's a test to see what's removed.
A maxdif of 1.9 looks a fair bit more untouched, I'd like to enable low by default for all content without it smoothing things, I can see the difference between a maxdif of 1.9 and 2.1 on my 65" plasma.
I have a hard time seeing any difference at all, and that is between the 2.1 and the "off" images.

i found a source which is hard banded on purpose:

avgDif maxDif grad Penalty LC Boost
average: 0.6 1.9 1.4 1.2 off 1.9

it's 32 mbit avg slow motion and some scene are not banded only the blue ones.

i don'f find a way to set up the max angle so this can be the problem too.

so i think it should be disabled on default.
You say "it" should be disabled on default. What is "it" in this context? And then you're talking about "the problem", but I'm not sure what problem you mean exactly. AFAIK in the original image there's banding in the blue sky and in the debanded image the banding is gone (and it should be). So what is "the problem" here?

I don't know about other people, but with the content I tested, these settings do very little. Actually they hardly do anything. And pretty much the only thing that works for bad content is increasing the boost and doing it significantly. How much? At the very least it should be 2.5. I have seen cases where there are bonuses for using a value of 4.0. I am sure we could find cases where even 5.5 would improve the picture.
Well, I watched your videos, there is a lot of mosquito noise and 8x8 MPEG2 blocks but not so much banding.

It's true that debanding can sometimes reduce those artefacts but it's not its main purpose. For that, madshi has to build a specific tool like deringing and deblocking.

For the medium preset, I tried to keep as much details as I could and to offer a real good debanding with medium quality content. If I increase the values too much, it blurs a lot of details. I could increase the angleBoost but then I had to decrease the other values. angleBoost doesn't affect the whole picture so I think it's a better idea to keep angleBoost around 1.9 and increase the other values so a good debanding is on even when angleBoost doesn't work.

For the high preset, I don't have a lot of very bad files. It could be higher and I think there is a good chance it will be.
Thanks for the samples, mikser. I do agree with Werewolfy, though, in that debanding is only meant to remove debanding. It's not meant to remove blocking, ringing or mosquito noise. Of course we can increase the parameters so that all of that is removed, too (together with all the details), but I don't think that's a good idea. There are better algorithms available which specifically target those other artifacts. We shouldn't misuse the debanding algorithm for that. You know, use the right tool for the right job etc.

It might make sense to at least increase the "high" preset a bit, though. But we've just started discussing it, so Werewolfy's suggestions are just that: A first suggestion.

does anyone know what nvidias G-Sync function is for?
AFAIK it only works for displays which support it (which is pretty much none today). I've no idea what effect that option has if the display doesn't support it.

I'm not sure that I'll be able to contribute to med/high settings until the weekend again, as things are quite busy for me right now.
No problem, we're not in a hurry. There will be no new official madVR release in the next 2 weeks or so, anyway.

One thing I have noticed though, is that debanding happens before scaling. This is normally fine, but can look quite bad with low resolution content.

I wonder if it would be possible to deband after scaling below a certain resolution? (or possibly at all resolutions?)
I don't see why debanding should be more effective or better after scaling. IMHO the best approach is to deband before scaling. Generally I believe that all algorithms which try to fix problems/artifacts in the source should normally run before scaling. Scaling by madVR is done in high bitdepth. So if the source is nicely debanded before scaling, it should still be nicely debanded afterwards.

If you think that debanding after scaling really works better, you could try upscaling with ffdshow raw video processor and then let madVR deband the ffdshow output. Of course ffdshow only scales in 8bit, so it's not ideal, but it might serve for some quick testing on your side...

And generally I think we should try to find values for high preset (ie. remove banding at all cost) and make medium with average values between low and high.
I have a specific definition for the 3 presets in mind:

Low: Highest priority on avoiding any detail loss, as far as possible, while still doing a useful amount of debanding.
Mid: Doing effective debanding for most sources, while accepting a very small (no more than that) loss of details.
High: Remove all debanding even from rather bad sources. Detail loss is acceptable, but no more than necessary.

I can live with the "High" preset being rather agressive, but it should still maintain a good "debanding vs detail" ratio. If that is not enough for some people, we might have to add an "Insane" preset, and I would actually name it that way.

I'd like "Low" and "Mid" both to be presets a user could use as default setting. The "Mid" setting would be for users who value banding-free images rather highly and who can live with a minor detail loss. While the "Low" preset would be for users who want to preserve as much detail as possible, but still want decent debanding at the same time.

As a result I expect "High" to be quite a bit higher than "Mid" and I expect "Mid" to be only moderately higher than "Low".

Maybe you could store presets in register? So everyone could adjust them.
I haven't even decided yet whether I will allow custom settings at all. If I decide to do that, I have not decided yet how to implement it. Of course the registry is one option, but there are others...

Yup and it's taken care of nicely with both settings. I don't mind if banded areas are smoother, but when what I consider to be non banded areas are changed, I really would like to see that minimized as much as possible, minor or not.

[...]

As you can see a maxdif of 2.1 encroaches more on non banded areas than with a maxdif of 1.9

Some numbers for comparison.

no deband vs low preset with maxdif 2.1

1510093 same pixel(s)
787040 unimportant difference pixel(s)
6867 important difference pixel(s)

no deband vs low preset with maxdif 1.9

1522985 same pixel(s)
774876 unimportant difference pixel(s)
6139 important difference pixel(s)

If it didn't affect debanding quality noticeably I'd push for a further maxdif reduction, but I suspect it would all start to add up.
What BeyondCompare does *not* tell you is how much debanding quality you're losing. And that's the key problem. What good is any statistic if it only shows you one side of the coin? Maybe with these test images a smaller "maxDif" looks every so slightly better to you (although I find it hard to see). But have you tested enough material? Maybe the small improvement is bought by drastic debanding loss in other videos? 8 eyes see more than 2. That's why I'm a bit worried about following the advice of one user who suggests other settings than 3 other users, who seem to more or less agree with each other.

That said, there's no harm in double checking your results. But without confirmation from Werewolfy, bacondither and 6233638 (or at least 2 of those 3) that the lowered "maxDif" threshold doesn't harm debanding I'm afraid of making things worse by changing it.

I have a Blu-Ray Player Sony S790 that has debanding. The low preset is very light, it preserves almost most details (but the debanding is weaker than our low preset :) ). The medium preset is just a bit stronger but the high preset is a lot stronger. It gets rid of almost all banding but the details are... destroyed :D
That is *extremely* interesting!!

I've heard that Sony offers a debanding feature in some of their devices and I wondered how effective it is. What is your impression about how it compares to what madVR does? You say madVR debands better at "low", but does the Sony have better detail preservation than madVR?

IIRC when you save image using Alt+I (Save Image function) madVR dithers on CPU and uses some kind of non-random dithering.
For Alt+I madVR uses some sort of error diffusion (similar to Floyd Steinberg), but it still uses a bit of random dithering, too. The result should be better quality than what madVR can (currently) do on the GPU.

Actually using Ctrl-I to save the image completely avoids debanding altogether. I'm getting bit identical output regardless of setting.
True. Currently screenshots are taken as early as possible in the rendering pipeline.

P.s. I've recently found video that is harmed by such "low" settings, sample (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/Rurouni_Kenshin_-_01.mkv).
No debanding (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_s.png), 1.9 (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_19.png), 2.1 (https://dl.dropboxusercontent.com/u/16254258/test/bandingfest/RK_21.png). Even there I don't really see any difference between 1.9 and 2.1.
So what is your opinion about our "low" preset parameters?

[6233638]

I don't see why debanding should be more effective or better after scaling. IMHO the best approach is to deband before scaling. Generally I believe that all algorithms which try to fix problems/artifacts in the source should normally run before scaling. Scaling by madVR is done in high bitdepth. So if the source is nicely debanded before scaling, it should still be nicely debanded afterwards.I agree entirely on the theory of processing at the native resolution. But when you have very low resolution content and are using a lot of debanding, the low resolution that dithering (debanding) is applied at can be very apparent. I wondered if it might be better to deband after scaling even if it meant that debanding itself was trickier.

When I get a chance to do more debanding testing, I'll see about installing ffdshow and giving that a try.

[madshi]

FWIW, madVR's debanding algorithm does not involve any dithering at all. All it does is smooth out gradients by calculating new pixel values in a much higher bitdepth. If dithering were involved, I might agree that debanding after scaling might help because then you wouldn't upscale dithering noise. But since dithering is not involved, I don't see how debanding after scaling would improve anything.

[huhn]

You say "it" should be disabled on default. What is "it" in this context? And then you're talking about "the problem", but I'm not sure what problem you mean exactly. AFAIK in the original image there's banding in the blue sky and in the debanded image the banding is gone (and it should be). So what is "the problem" here?


the source is very clean except the ending cause they want to it to look like that.

it does a good job here in a normal scene and this a very hard scene too. but it make's it a little bit unsharp
http://abload.de/thumb/shincleanloweqs3y.png (http://abload.de/image.php?img=shincleanloweqs3y.png)http://abload.de/thumb/shincleanoffsisbv.png (http://abload.de/image.php?img=shincleanoffsisbv.png)

but this here is a hole different story
i'm very sure it is there because they what is to look like this the ending feels like a nightmare. it's like an unsharp dream scene just with banding.
http://abload.de/thumb/shinedlowaqscr.png (http://abload.de/image.php?img=shinedlowaqscr.png)http://abload.de/thumb/shinedoffkqsuc.png (http://abload.de/image.php?img=shinedoffkqsuc.png)

[Ver Greeneyes]

but this here is a hole different story
i'm very sure it is there because they what is to look like this the ending feels like a nightmare. it's like an unsharp dream scene just with banding.
http://abload.de/thumb/shinedlowaqscr.png (http://abload.de/image.php?img=shinedlowaqscr.png)http://abload.de/thumb/shinedoffkqsuc.png (http://abload.de/image.php?img=shinedoffkqsuc.png)
To be honest I don't feel the debanding detracts from the scene there. It's visibly different, yes, but worse? I think it looks better. Of course, that's another reason why having multiple presets is useful - not everyone is going to want the same thing.

[madshi]

the source is very clean except the ending cause they want to it to look like that.

it does a good job here in a normal scene and this a very hard scene too. but it make's it a little bit unsharp
http://abload.de/thumb/shincleanloweqs3y.png (http://abload.de/image.php?img=shincleanloweqs3y.png)http://abload.de/thumb/shincleanoffsisbv.png (http://abload.de/image.php?img=shincleanoffsisbv.png)

but this here is a hole different story
i'm very sure it is there because they what is to look like this the ending feels like a nightmare. it's like an unsharp dream scene just with banding.
http://abload.de/thumb/shinedlowaqscr.png (http://abload.de/image.php?img=shinedlowaqscr.png)http://abload.de/thumb/shinedoffkqsuc.png (http://abload.de/image.php?img=shinedoffkqsuc.png)
Again: You said earlier that "it" should be disabled on default. What did you mean with "it"? Please try to be a bit clearer. Did you mean *debanding* should be disabled by default? Or the "angleBoost" parameter? Do you mean we should change the "low" parameters to do less debanding? Or what is it that you want to say exactly?

FWIW, I do not plan to enable debanding by default. The reason for that is performance, though. I don't want low level GPUs to stutter because I've enabled debanding by default. I will probably recommend to enable debanding and set it to "low", though, for all users if the GPU is fast enough.

About your comparison images: I'm not sure myself whether these are banding artifacts that should be removed or not. On a quick check I would say they are and they should. But it's probably only possible to judge when seen in motion. "Intended banding" would probably stay stable in motion, while an unintended banding artifact likes to move around. But that's not something the debanding algorithm can detect. FWIW, if I were in the shoes of the algorithm, when looking at your screenshots, I'd say those are probably banding artifacts that should be removed.

[huhn]

is was about the debanding filter not a setting or something.

i don't like postprogressing by default but if this:
FWIW, I do not plan to enable debanding by default. is the case i think i'm totally fine with it.

i through you where going to enable low by default my fail.

"Intended banding" would probably stay stable in motion,

it is stable totally

there is a scene blending which is not stable but 35 mbit for this type of content is my be not enough.

[Werewolfy]

That is *extremely* interesting!!

I've heard that Sony offers a debanding feature in some of their devices and I wondered how effective it is. What is your impression about how it compares to what madVR does? You say madVR debands better at "low", but does the Sony have better detail preservation than madVR?


I have a LED TV Sony W905 and a Blu-Ray player BDP-S790, both have a debanding feature. On the Blu-Ray player with Super Bit Mapping activated and Deep Color 12 bit, the low preset has much less detail preservation and a weaker debanding strength than madVR ;) Detail preservation is similar to my current medium preset and debanding strength is similar to my "very low" preset both posted earlier.

On the TV, the low preset is really too strong, between the medium and high preset of the Blu-Ray Player.

[madshi]

there is a scene blending which is not stable but 35 mbit for this type of content is my be not enough.
35mbit sounds like a lot to me, especially for anime content. I think the max video bitrate for Blu-Ray is 40mbit, IIRC. Anime usually needs less video bitrate than normal movie content, I think.

I have a LED TV Sony W905 and a Blu-Ray player BDP-S790, both have a debanding feature. On the Blu-Ray player with Super Bit Mapping activated and Deep Color 12 bit, the low preset has much less detail preservation and a weaker debanding strength than madVR ;) Detail preservation is similar to my current medium preset and debanding strength is similar to my "very low" preset both posted earlier.

On the TV, the low preset is really too strong, between the medium and high preset of the Blu-Ray Player.
Thank you. That's exactly what I wanted to hear... :D:D

[huhn]

35mbit sounds like a lot to me, especially for anime content. I think the max video bitrate for Blu-Ray is 40mbit, IIRC.

even on 40 mbit vbr disc there is still bending so.

nearly all anime disc today have more than 32 mbit

and yeah 40 mbit vbr is max and 48 mbit is cap for all streams on 2d disc.

Anime usually needs less video bitrate than normal movie content, I think.

i hear that a lot but that doesn't mean 40 mbit is enough even for anime.

[ryrynz]

What BeyondCompare does *not* tell you is how much debanding quality you're losing.


I do realize this is a two sided coin and that's exactly the reason I said "if it didn't affect debanding quality, I'd reduce it further"
I haven't looked at lower maxdif values but nobody has claimed that debanding is affected greatly, Werewolfy has mentioned he's seen some light changes but didn't see any negative results dropping maxdif.


That said, there's no harm in double checking your results. But without confirmation from Werewolfy, bacondither and 6233638 (or at least 2 of those 3) that the lowered "maxDif" threshold doesn't harm debanding I'm afraid of making things worse by changing it.


From what I've seen lowering the maxdif to 1.9 doesn't harm debanding much, could you guys look at lowering maxdif levels to 1.9 & also 1.8 and compare that vs 2.1?
The main focus here is to avoid changing pixels that really don't need to be changed. If we can get the same debanding quality to our eyes by changing less pixels, that IMO should be preferred.
The low preset should be for maintaining as much of the original image as possible without visibly reducing deband quality from what we've established to be effective levels for this preset.

Madshi are there any more tweaks forthcoming for debanding? I know you were looking at some of 6233638's material for possibly better banding detection.

[THX-UltraII]

I don t want to ruin the discussion that is about the debanding etc that seems to be pretty hot atm but I m still struggeling with my setup:

2 weeks ago I switched from my Intel HD2000 graphics card to the HD6950. 2 most important reasons for this where 1. the Intel HD Graphics cards have problems with RGB Full Range output and 2. Intel HD2000 Graphics does not have enough horse power for my purpose which is 2K=>4K upscaling. I want to 2K=4K upscale because I want to feed my native 4K projector with a native 4K signal and let my HTPC do all the scaling.

But I found out that the HD6950 max resolution over HDMI is 1920x1080. So I swapped it for a HD7950. According to specs the HD7950 supports 4096x2160 over HDMI. So I installed the card last night but the AMD control panel says that the max supported resolution is 1080p. Could my Onkyo TX-NR1010 receiver be the problem here?

[LigH]

If you pass the HDMI output through your A/V receiver, it can indeed limit the supported resolutions. It has to forward the monitor's supported resolutions and thus needs to understand them to some degree. I remember having a little trouble already with a 1920x1200 monitor connected via DVI adapter...

[THX-UltraII]

If you pass the HDMI output through your A/V receiver, it can indeed limit the supported resolutions. It has to forward the monitor's supported resolutions and thus needs to understand them to some degree. I remember having a little trouble already with a 1920x1200 monitor connected via DVI adapter...So I just found out that the Onkyo TX-NR1010 does not support 4K passthrough. Going to switch my AVR to a 4K passthrough compatible AVR like the Marantz AV7008.

After I get the right AVR, how do I exactly config my HTPC? I can understand that I will need to output 4096x2160 native via the AMD control panel but Madshi told me something about 'let MPC-BE zoom'. What does this mean?

[leeperry]

I'm just not sure if it makes any sense for madVR to offer BT.1886
Actually, having more thoughts about it I would thoroughly enjoy an option for BT.1886 gamma in mVR as I think I will:
-run that TV in PC mode, which disables its gamma setting altogether and allows for 4:4:4 display
-calibrate it to D65 using its R/G/B offset/gain settings using HCFR, which is child's play
-roll gamuts with automatic PotP profiles and that PS gamut script, child's play again
-mess with gamma in mVR in 16bit

The nice thing about all this is that I'll neither have to bother with 3DLUT's or 8bit CLUT's, but I gotta admit that BT.1886 would be the icing on the cake :o

[madshi]

even on 40 mbit vbr disc there is still bending so.

i hear that a lot but that doesn't mean 40 mbit is enough even for anime.
I think it's not the bitrate. I think the banding is either already in the studio master, or if not, then maybe the encoding house failed to apply proper dithering before encoding. Anyway, makes no difference to us.

are there any more tweaks forthcoming for debanding? I know you were looking at some of 6233638's material for possibly better banding detection.
No further tweaks planned. Ok, maybe one little thing, but I expect that it will not result in any changes to any of the current parameters, so I'll wait until low, medium and high presets are decided on.

So I just found out that the Onkyo TX-NR1010 does not support 4K passthrough. Going to switch my AVR to a 4K passthrough compatible AVR like the Marantz AV7008.
I would wait with that. First check whether letting madVR upscale to 4K actually produces better results than letting your projector do that. The Sony upscaling algorithm isn't so bad, so it's worth a try which looks better to your eyes. If you prefer the Sony algorithm, you can stick to the Onkyo. You can compare the algorithms right now by simply bypassing the Onkyo. Connect your HTPC directly to the Sony. You'll have no sound this way, but you don't need it to judge image quality, obviously. Simply compare how the image quality looks like with your GPU set to either 1080p output or 4K output. Pick the configuration which looks better to your eyes.

Madshi told me something about 'let MPC-BE zoom'.
menu -> View -> Video Frame -> Touch Window From Inside

Come on, THX-UltraII, these are beginner questions. It would have cost you like 10 seconds to find the solution for this yourself. You could save me quite a bit of time by simply trying some things for yourself instead of asking every tiny bit of information.

[THX-UltraII]

I would wait with that. First check whether letting madVR upscale to 4K actually produces better results than letting your projector do that. The Sony upscaling algorithm isn't so bad, so it's worth a try which looks better to your eyes. If you prefer the Sony algorithm, you can stick to the Onkyo. You can compare the algorithms right now by simply bypassing the Onkyo. Connect your HTPC directly to the Sony. You'll have no sound this way, but you don't need it to judge image quality, obviously. Simply compare how the image quality looks like with your GPU set to either 1080p output or 4K output. Pick the configuration which looks better to your eyes.We talked a little bit about this before. I only think about this without any further knowlegde but if I let the Sony do the 4K upscaling the picture is 'adjusted' more times: first madVR does chroma upscaling (I use JINC8+AR). After this the video is ouputted in 1080p to the Sony and then the Sony does its 2K=>4K upscaling thing. I thought you told me it is better to let madVR do all the scaling (I think luma scaling is activated when madVR does the 2K>4K scaling right?) and output in native 4K to the Sony.


menu -> View -> Video Frame -> Touch Window From Inside

Come on, THX-UltraII, these are beginner questions. It would have cost you like 10 seconds to find the solution for this yourself. You could save me quite a bit of time by simply trying some things for yourself instead of asking every tiny bit of informationSorry madshi. I really did not have a clue what you meant......

[madshi]

We talked a little bit about this before. I only think about this without any further knowlegde but if I let the Sony do the 4K upscaling the picture is 'adjusted' more times: first madVR does chroma upscaling (I use JINC8+AR). After this the video is ouputted in 1080p to the Sony and then the Sony does its 2K=>4K upscaling thing. I thought you told me it is better to let madVR do all the scaling (I think luma scaling is activated when madVR does the 2K>4K scaling right?) and output in native 4K to the Sony.
I've told you in my comment #20547 what I would do in your place. Either follow my suggestion or don't. But I won't discuss with you what you think I might have said in the past or not. And I won't answer any further questions about this topic, unless you've followed my suggestion first.

[THX-UltraII]

Ok, I will try as you suugested first Madshi.

But the reason for asking this again was your post #20340.

Usually the best solution is to let your HTPC do all the processing and send the final renderer video image in the native resolution of your projector (4K) to the projector.

This is the reason why I bought the HD7950 and also the reason why I m thinking about buying a 4K passthrough capable AVR

[madshi]

"Usually" is not the same as "always". My reply #20340 was mostly a technical answer to your question about whether source devices should output video untouched or not. "Usually" you get better results by letting madVR scale. But that doesn't always have to be the case. I've not seen the Sony scaling algorithm myself, so I can't judge that. It's always the best idea to compare all options and choose the one which looks best to your eyes. Why would you trust me when you can find out for yourself by simply trying and comparing, and when I've not seen your projector myself yet? You're in a much better position to find out which configuration looks best than I am.

[THX-UltraII]

"Usually" is not the same as "always". It doesn't hurt to compare. I always say "trust your eyes".

Ok, I will test and report back.

I almost don t even dare to ask other questions anymore but could you tell me what the difference in PQ will be when using DXVA (HW) decoding and when NOT using any HW decoding (software decoding?).

I ask this because I have a fast CPU. If I can my CPU do the video decoding I will have more available power from my GPU for the most GPU intensive algorithms. But if there are any downsides of using software decoding I won t do it of course.

[Werewolfy]

I do realize this is a two sided coin and that's exactly the reason I said "if it didn't affect debanding quality, I'd reduce it further"
I haven't looked at lower maxdif values but nobody has claimed that debanding is affected greatly, Werewolfy has mentioned he's seen some light changes but didn't see any negative results dropping maxdif.

From what I've seen lowering the maxdif to 1.9 doesn't harm debanding much, could you guys look at lowering maxdif levels to 1.9 & also 1.8 and compare that vs 2.1?
The main focus here is to avoid changing pixels that really don't need to be changed. If we can get the same debanding quality to our eyes by changing less pixels, that IMO should be preferred.
The low preset should be for maintaining as much of the original image as possible without visibly reducing deband quality from what we've established to be effective levels for this preset.


Between maxDif 1.9 and 2.1 there is a slight difference in debanding quality, I find 1.9 is just a little too low sometimes. MaxDif at 2.0 could be a better compromise, I don't think there is a difference between 2.0 and 2.1. But it must be confirmed by other testers.

[pie1394]

<skip...>
Connect your HTPC directly to the Sony. You'll have no sound this way, but you don't need it to judge image quality, obviously. Simply compare how the image quality looks like with your GPU set to either 1080p output or 4K output. Pick the configuration which looks better to your eyes.


If I don't remember it wrong, the CCC allows us to select any HDMI port as the default audio output. So it might be possible to make an audio-only link to AV receiver via another HDMI port...

But there is no HDMI input on my old-fashioned AV receiver (Yamaha RX-V1500). So I cannot make a test to verify that with my HD7970. :p

Recently I just also noticed that my ION system's cooling fan sounds louder frequently. After some checking, I have found the root cause is from driver 327.23. The shader clock keeps locked at 800 MHz at idle... Go back to driver 320.49, and it can go down as low as 400MHz again.