Welcome to Doom9's Forum, THE in-place to be for everyone interested in DVD conversion.

Before you start posting please read the forum rules. By posting to this forum you agree to abide by the rules.

 

Go Back   Doom9's Forum > Hardware & Software > Software players

Reply
 
Thread Tools Search this Thread Display Modes
Old 11th May 2015, 20:28   #29681  |  Link
James Freeman
Registered User
 
Join Date: Sep 2013
Posts: 919
Quote:
Originally Posted by madshi View Post
Was a stupid bug on my side, which seemingly Windows 8.1 x64 didn't have a problem with. My best guess is that Windows 7 didn't like it.
Nope.
W7 64 over here, and the first 88.3 worked fine.
__________________
System: i7 3770K, GTX660, Win7 64bit, Panasonic ST60, Dell U2410.
James Freeman is offline   Reply With Quote
Old 11th May 2015, 20:28   #29682  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
madVR v0.88.4 released

http://madshi.net/madVR.zip

Code:
* fixed startup crash in Windows 7, introduced in v0.88.3
* NNEDI3 doubling: chroma is now always upscaled with max 2 taps
madshi is offline   Reply With Quote
Old 11th May 2015, 20:49   #29683  |  Link
Razoola
Registered User
 
Join Date: May 2007
Posts: 454
088.4 has seemed to fix a lot of issues I was having with MPC-HC not closing down correctly
Razoola is offline   Reply With Quote
Old 11th May 2015, 20:50   #29684  |  Link
iSunrise
Registered User
 
Join Date: Dec 2008
Posts: 496
Quote:
Originally Posted by madshi View Post
* NNEDI3 doubling: chroma is now always upscaled with max 2 taps
Can you please explain the last two changes regarding NNEDI3 chroma doubling in a bit more detail?

I guess you removed NNEDI3 chroma doubling, because there really was no use case for it and no one could see differences?* What exactly is the second change, I am not sure I understand it and what your intention with that change is?

Thanks!

*I think I remember that chroma upscaling was used first anyway and when you do image doubling afterwards, you basically already have a fully upsampled chroma so the whole image (chroma+luma) is doubled, so no need for chroma doubling anymore. Correct?

Last edited by iSunrise; 11th May 2015 at 20:55.
iSunrise is offline   Reply With Quote
Old 11th May 2015, 20:53   #29685  |  Link
yok833
Registered User
 
Join Date: Aug 2012
Posts: 73
Quote:
Originally Posted by madshi View Post
madVR v0.88.4 released

http://madshi.net/madVR.zip

Code:
* fixed startup crash in Windows 7, introduced in v0.88.3
* NNEDI3 doubling: chroma is now always upscaled with max 2 taps
Many thanks Madshi.... I dream that one day the developers from nvidia fix their firmwares as fast as you do
yok833 is offline   Reply With Quote
Old 11th May 2015, 21:25   #29686  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by iSunrise View Post
Can you please explain the last two changes regarding NNEDI3 chroma doubling in a bit more detail?

I guess you removed NNEDI3 chroma doubling, because there really was no use case for it and no one could see differences?* What exactly is the second change, I am not sure I understand it and what your intention with that change is?
When doing image doubling, the "input" into the doubling algorithm is always RGB. When using NNEDI3 for doubling, this RGB image is then split into Luma and Chroma. Luma is doubled with NNEDI3. Chroma is doubled by using the "image upscaling" algorithm - but only if it has less than 3 taps. If the "image upscaling" algorithm has 3 taps or more, madVR doubles Chroma by using Catmull-Rom instead. I've implemented these changes because the algorithm used for Chroma doubling is very very unimportant, so it's no use wasting precious GPU processing time there.
madshi is offline   Reply With Quote
Old 11th May 2015, 21:30   #29687  |  Link
iSunrise
Registered User
 
Join Date: Dec 2008
Posts: 496
Wow! I cannot believe it. This is just great.

Even though all information I could find about firmware and internals said otherwise, my Eizo CG243W does indeed accept full 10bit input over a single-link DVI-D connection (it doesn't have dual-link).

For anyone that is interested, I am testing with LightIllusion's official calibration images (.TIF) in 16bit (48bit) precision, where I can clearly discern differences between 8bit and 10bit. And I didn't expect that I could even tell them apart, but I clearly can. The pattern is completely smooth, like a baby's arse (so to speak). Huge improvement compared to 8bit. Didn't expect that at all.

Finally I am able to use my expensive CG also for maximum precision movie playback! Perfect!

Thank you so much madshi for creating madVR, this is another of these times when I am completely astounded by your work. Just...genius.

One very big important thing though:
It only works in FSE mode, so don't get fooled when it doesn't work in windowed, you have to be in FSE.

Last edited by iSunrise; 11th May 2015 at 21:33.
iSunrise is offline   Reply With Quote
Old 11th May 2015, 21:32   #29688  |  Link
Anima123
Registered User
 
Join Date: Jun 2005
Posts: 504
Quote:
Originally Posted by madshi View Post
madVR v0.88.4 released

http://madshi.net/madVR.zip

Code:
* fixed startup crash in Windows 7, introduced in v0.88.3
* NNEDI3 doubling: chroma is now always upscaled with max 2 taps
Can someone confirm that the dithering of Error Diffusion not working anymore? I mostly judge by the rendering time not increasing when enabled, unlike before.
Anima123 is offline   Reply With Quote
Old 11th May 2015, 21:35   #29689  |  Link
iSunrise
Registered User
 
Join Date: Dec 2008
Posts: 496
Quote:
Originally Posted by Anima123 View Post
Can someone confirm that the dithering of Error Diffusion not working anymore? I mostly judge by the rendering time not increasing when enabled, unlike before.
Works just fine for me.

You can easily test it if you go to the display's native bitdepth setting and set something very low like 2bit. When you enable any of the dithering options they all work as intended, you get a lot more precision as a result (lots of dots).
iSunrise is offline   Reply With Quote
Old 11th May 2015, 21:44   #29690  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by iSunrise View Post
Wow! I cannot believe it. This is just great.

Even though all information I could find about firmware and internals said otherwise, my Eizo CG243W does indeed accept full 10bit input over a single-link DVI-D connection (it doesn't have dual-link).

For anyone that is interested, I am testing with LightIllusion's official calibration images (.TIF) in 16bit (48bit) precision, where I can clearly discern differences between 8bit and 10bit. And I didn't expect that I could even tell them apart, but I clearly can. The pattern is completely smooth, like a baby's arse (so to speak). Huge improvement compared to 8bit. Didn't expect that at all.
Is that with dithering enabled or disabled? FWIW, if 10bit works correctly from start to finish, if you enable dithering, you should not see any difference in smoothness, but you should see a lower noise floor in 10bit compared to 8bit. Is that what you're seeing?

Quote:
Originally Posted by Anima123 View Post
Can someone confirm that the dithering of Error Diffusion not working anymore? I mostly judge by the rendering time not increasing when enabled, unlike before.
It's working fine here. 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.
madshi is offline   Reply With Quote
Old 11th May 2015, 21:55   #29691  |  Link
JarrettH
Registered User
 
Join Date: Aug 2004
Location: Canada
Posts: 860
Quote:
Originally Posted by madshi View Post
Try Lanczos3 AR instead. Or even Catmull Rom AA. If you're using image doubling + quadrupling, the image upscaling algorithm is much less important, and it's used for upscaling the chroma part of the image, when using NNEDI3 for image doubling.
Could you expand on that? What effect does the choice of luma upscaling have on image doubling? Is only chroma upscaling happening?

What happens with image doubling?
If my video is 1280x720, does it then become 2560x1440, then downscaled to fit my 1080p screen?

JarrettH is offline   Reply With Quote
Old 11th May 2015, 21:58   #29692  |  Link
iSunrise
Registered User
 
Join Date: Dec 2008
Posts: 496
Quote:
Originally Posted by madshi View Post
Is that with dithering enabled or disabled? FWIW, if 10bit works correctly from start to finish, if you enable dithering, you should not see any difference in smoothness, but you should see a lower noise floor in 10bit compared to 8bit. Is that what you're seeing?
I basically disabled dithering and worked my way up from 2bit to fnally 10bit, where the important step from 8bit to 10bit is making a huge difference. When I am at 8bit I can still see very tiny "lines" from top to bottom [with the described 16bit (48bit) BWramp] and with 10bit I cannot discern any gradient steps anymore with my eyes, even when my nose is glued to the display. It's like a B&W rainbow with too many steps that I cannot make out anymore. Completely smooth.

About the additional dithering, yes, that should indeed lead to a lower noise floor, since your output is fp16, right? When judging purely with my eyes, though, 10bit seems to be the absolute limit that I can personally perceive (on this display at least). The additional dithering may just be the icing on the cake.

Last edited by iSunrise; 11th May 2015 at 22:07.
iSunrise is offline   Reply With Quote
Old 11th May 2015, 22:05   #29693  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
Quote:
Originally Posted by JarrettH View Post
What happens with image doubling?
If my video is 1280x720, does it then become 2560x1440, then downscaled to fit my 1080p screen?

that's how image doubling works. and the reason i think 1.2x or always are a total waste.
huhn is offline   Reply With Quote
Old 11th May 2015, 22:11   #29694  |  Link
Asmodian
Registered User
 
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,406
Quote:
Originally Posted by JarrettH View Post
Could you expand on that? What effect does the choice of luma upscaling have on image doubling? Is only chroma upscaling happening?

What happens with image doubling?
If my video is 1280x720, does it then become 2560x1440, then downscaled to fit my 1080p screen?
In your example:

1280x720 luma is doubled to 2560x1440 and then scaled to 1920x1080 with image downscaling.
1280x720 chroma is scaled to 1920x1080 with image upscaling.
__________________
madVR options explained
Asmodian is offline   Reply With Quote
Old 11th May 2015, 22:25   #29695  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Quote:
Originally Posted by iSunrise View Post
I basically disabled dithering and worked my way up from 2bit to fnally 10bit, where the important step from 8bit to 10bit is making a huge difference. When I am at 8bit I can still see very tiny "lines" from top to bottom [with the described 16bit (48bit) BWramp] and with 10bit I cannot discern any gradient steps anymore with my eyes, even when my nose is glued to the display. It's like a B&W rainbow with too many steps that I cannot make out anymore. Completely smooth.

About the additional dithering, yes, that should indeed lead to a lower noise floor, since your output is fp16, right? When judging purely with my eyes, though, 10bit seems to be the absolute limit that I can personally perceive (on this display at least). The additional dithering may just be the icing on the cake.
Ok, disabling dithering is a valid test, but only for checking if there's any difference between 8bit and 10bit output at all. You should always have dithering enabled in real life. So that's what you should test with, in order to decide whether using 10bit output has any real benefit. So I would recommend to enable dithering again and then check if you can see any difference in image quality between 8bit and 10bit.
madshi is offline   Reply With Quote
Old 11th May 2015, 22:39   #29696  |  Link
Moony349
Registered User
 
Join Date: Sep 2012
Posts: 25
Quote:
Originally Posted by madshi View Post
When doing image doubling, the "input" into the doubling algorithm is always RGB. When using NNEDI3 for doubling, this RGB image is then split into Luma and Chroma. Luma is doubled with NNEDI3. Chroma is doubled by using the "image upscaling" algorithm - but only if it has less than 3 taps. If the "image upscaling" algorithm has 3 taps or more, madVR doubles Chroma by using Catmull-Rom instead. I've implemented these changes because the algorithm used for Chroma doubling is very very unimportant, so it's no use wasting precious GPU processing time there.


Hi madshi,

So if I understand correctly, we should ignore the chroma > Jinc3 AR above and, going forward, MadVR defaults to Catmull-Rom for chroma if NNEDI3 is used for luma?
Moony349 is offline   Reply With Quote
Old 11th May 2015, 22:41   #29697  |  Link
Anima123
Registered User
 
Join Date: Jun 2005
Posts: 504
Quote:
Originally Posted by madshi View Post
Is that with dithering enabled or disabled? FWIW, if 10bit works correctly from start to finish, if you enable dithering, you should not see any difference in smoothness, but you should see a lower noise floor in 10bit compared to 8bit. Is that what you're seeing?
According to the GPU usage, I am afraid dithering fall back to ordered dithering all the time even with 8-bit output. Could you please double check?

I didn't encounter such problem with previous versions.
Anima123 is offline   Reply With Quote
Old 11th May 2015, 22:43   #29698  |  Link
nevcairiel
Registered Developer
 
Join Date: Mar 2010
Location: Hamburg/Germany
Posts: 10,342
Quote:
Originally Posted by Moony349 View Post
So if I understand correctly, we should ignore the chroma > Jinc3 AR above and, going forward, MadVR defaults to Catmull-Rom for chroma if NNEDI3 is used for luma?
It even tells you this in the second Chroma line.

First, 4:2:0 Chroma is upscaled to 4:4:4/RGB using your selected chroma upscaling algorithm. Then image doubling is used for Luma, and the chroma is upscaled further using Catmull-Rom AR.
__________________
LAV Filters - open source ffmpeg based media splitter and decoders
nevcairiel is offline   Reply With Quote
Old 11th May 2015, 22:44   #29699  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,812
I found out that my problem with NNEDI3 + FineSharp afterwards is caused by image quadrupling.
Doubling 64 + FS works fine, but additional quadroupling 64 produces corruption and NV driver failure.
Anyone confirm? Be sure that quadrupling is actually active.
aufkrawall is offline   Reply With Quote
Old 11th May 2015, 22:58   #29700  |  Link
Hprd
Registered User
 
Join Date: Mar 2011
Posts: 60
Quote:
Originally Posted by aufkrawall View Post
I found out that my problem with NNEDI3 + FineSharp afterwards is caused by image quadrupling.
Doubling 64 + FS works fine, but additional quadroupling 64 produces corruption and NV driver failure.
Anyone confirm? Be sure that quadrupling is actually active.
Finesharp (and luma sharp) works fine with quadrupling, it's super res that crashes in this situation for me, lol
Hprd is offline   Reply With Quote
Reply

Tags
direct compute, dithering, error diffusion, madvr, ngu, nnedi3, quality, renderer, scaling, uhd upscaling, upsampling

Thread Tools Search this Thread
Search this Thread:

Advanced Search
Display Modes

Posting Rules
You may not post new threads
You may not post replies
You may not post attachments
You may not edit your posts

BB code is On
Smilies are On
[IMG] code is On
HTML code is Off

Forum Jump


All times are GMT +1. The time now is 14:05.


Powered by vBulletin® Version 3.8.11
Copyright ©2000 - 2024, vBulletin Solutions Inc.