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 6th July 2015, 00:06   #31541  |  Link
Anima123
Registered User
 
Join Date: Jun 2005
Posts: 504
Quote:
So I can remove super-xbr-25 again?
Madshi, please keep super-xbr-25 option, which could be useful for some not well coded videos. super-xbr-25 would be sharp enough but not overly sharped.

For Chroma upscaling, shouldn't we use a less sharper algorithm? In this case, super-xbr-25 can also be useful.

Anyone ever tried with super-xbr-25 for chroma upscaling?
Anima123 is offline   Reply With Quote
Old 6th July 2015, 04:05   #31542  |  Link
Catclaw
Registered User
 
Join Date: Jun 2015
Posts: 3
Any idea why using CUVID or NNEDI3 chroma upscaling crashes windows 10? and is any progress being made towards madvr on windows 10?
Catclaw is offline   Reply With Quote
Old 6th July 2015, 05:24   #31543  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by Catclaw View Post
Any idea why using CUVID or NNEDI3 chroma upscaling crashes windows 10? and is any progress being made towards madvr on windows 10?
Until it's released its almost pointless looking at it. The code changes daily..
ryrynz is offline   Reply With Quote
Old 6th July 2015, 09:21   #31544  |  Link
viewer
Registered User
 
Join Date: May 2015
Posts: 5
i prefer super-xbr strict AR. super-xbr high-quality AR is something ringing.


viewer is offline   Reply With Quote
Old 6th July 2015, 12:45   #31545  |  Link
xabregas
Registered User
 
Join Date: Jun 2011
Posts: 121
Hi,

I must say super-xbr is amazing, but why not put it in luma upscaling below jinc? I found every luma upscaling algorithm very weak compared to luma doubling even for 720p > 1080p upscaling. Even knowing it will upscale 2 x 720p and then downscale to 1080p using bicubic or wtv i prefer using luma double because i feel it has better PQ to my eyes. So, why not put that option in luma upscaling? It shouldnt be much gpu intensive given the performance boost we get from NNEDI3 to super-xbr. Meaning, using super-xbr to 720p -> 1080p upscale.
xabregas is offline   Reply With Quote
Old 6th July 2015, 13:14   #31546  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by viewer View Post
i prefer super-xbr strict AR. super-xbr high-quality AR is something ringing.
Woah, that's messed up! Look at all those green lines.. Madshi, any room for improvement there?
ryrynz is offline   Reply With Quote
Old 6th July 2015, 14:56   #31547  |  Link
michkrol
Registered User
 
Join Date: Nov 2012
Posts: 167
Quote:
Originally Posted by madshi View Post
* modified Bilateral chroma upscaling algorithm parameters
Thanks for the new version of madVR, especially this.

After some testing, Bilateral is now my chroma upscaling algorithm for anime.
It's pure magic, especially with high quality sources

Quote:
Originally Posted by viewer View Post
i prefer super-xbr strict AR. super-xbr high-quality AR is something ringing.
You should give Bilateral a try. I really like the results with high quality sources.
Your sample with bilateral chroma scaling below.


Quote:
Originally Posted by xabregas View Post
Hi,

I must say super-xbr is amazing, but why not put it in luma upscaling below jinc? I found every luma upscaling algorithm very weak compared to luma doubling even for 720p > 1080p upscaling. Even knowing it will upscale 2 x 720p and then downscale to 1080p using bicubic or wtv i prefer using luma double because i feel it has better PQ to my eyes. So, why not put that option in luma upscaling? It shouldnt be much gpu intensive given the performance boost we get from NNEDI3 to super-xbr. Meaning, using super-xbr to 720p -> 1080p upscale.
I don't understand you reasoning behind this.

Biggest questions: what would madVR do if you selected s-xbr upscaling and nnedi3 doubling? What image scaling algo would madVR use after doubling with s-xbr?
It's under doubling because you can only double, not scale by any (arbitrary) ratio, performance has nothing to do with this. The up-/down-scaling algo used after doubling is important for image quality.

Last edited by michkrol; 6th July 2015 at 16:03. Reason: wording
michkrol is offline   Reply With Quote
Old 6th July 2015, 15:20   #31548  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,812
Quote:
Originally Posted by Catclaw View Post
Any idea why using CUVID or NNEDI3 chroma upscaling crashes windows 10? and is any progress being made towards madvr on windows 10?

Cuvid + needi3 chroma scaling is broken here (GTX 980, 353.38, Windows 10 10162 x64) too, but NNEDI3 chroma scaling only works well (e.g. with DXVA2 CB).
madVR is running very well here on Windows 10 apart from that unimportant Cuvid issue.
aufkrawall is offline   Reply With Quote
Old 6th July 2015, 16:43   #31549  |  Link
Vyral
Registered User
 
Vyral's Avatar
 
Join Date: Oct 2012
Posts: 70
Quote:
Originally Posted by michkrol View Post
After some testing, Bilateral is now my chroma upscaling algorithm for anime.
It's pure magic, especially with high quality sources
Really !?

I didn't even try this algorithm because many users here said it was horrible.
__________________
iiyama prolite xb2483hsu 1080p60 Gamma=2.25 - Intel Core i3-2100 3.10GHz - AMD Radeon HD 6850, RGB 4:4:4 Full range - MPC-HC + XYSubFilter + madVR
Vyral is offline   Reply With Quote
Old 6th July 2015, 17:01   #31550  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,812
With lowres sources, heavy color bleeding can be observed.

Jinc AR:


Bileteral:
aufkrawall is offline   Reply With Quote
Old 6th July 2015, 17:18   #31551  |  Link
omarank
Registered User
 
Join Date: Nov 2011
Posts: 187
Quote:
Originally Posted by madshi View Post
It's still a 4K image. On a quick check I didn't know where to look. I did a quick check but didn't see any obvious differences between Jinc or super-xbr chroma upscaling.
The differences between the two chroma upscaling algorithms on a still image like this may not be obvious. You may look at the trees or the mountains in the background. When Jinc chroma is used, there is a natural depth to the image, while when super-xbr chroma is used, the image loses that depth. Well this is how I perceive it. And these differences are more observable when mono dithering is used.

Quote:
Originally Posted by madshi View Post
Thanks! You were using strength=1.0 and softness=0.0, right? I'm asking just to be sure, because madVR did store and reuse the settings you used from the older SuperRes.
Right.
omarank is offline   Reply With Quote
Old 6th July 2015, 18:54   #31552  |  Link
chros
Registered User
 
chros's Avatar
 
Join Date: Mar 2002
Posts: 2,323
Quote:
super-xbr chroma upscaling now supports higher quality AR algo -> 6% slower
Quote:
When Jinc chroma is used, there is a natural depth to the image, while when super-xbr chroma is used, the image loses that depth.
The new chroma upscaling super-xbr + HQ AR is slower now than Jinc + AR , so it's a no go for me (tested on my 650m in p5 powerstate to clearly see the difference).

And 1 more interesting thing: D3D9 exclusive Old path is way-way faster (~40%) then the New path with the same settings (when you switch on several frames in advance)! Is there any advantage (quality wise) to use the New path if everything else seems to work fine with the old one?

Edit: and D3D11 exclusive is the slowest from all them.
__________________
Ryzen 5 2600,Asus Prime b450-Plus,16GB,MSI GTX 1060 Gaming X 6GB(v398.18),Win10 LTSC 1809,MPC-BEx64+LAV+MadVR,Yamaha RX-A870,LG OLED77G2(2160p@23/24/25/29/30/50/59/60Hz) | madvr config

Last edited by chros; 6th July 2015 at 19:16.
chros is offline   Reply With Quote
Old 6th July 2015, 19:39   #31553  |  Link
xabregas
Registered User
 
Join Date: Jun 2011
Posts: 121
Quote:
Originally Posted by michkrol View Post
I don't understand you reasoning behind this.

Biggest questions: what would madVR do if you selected s-xbr upscaling and nnedi3 doubling? What image scaling algo would madVR use after doubling with s-xbr?
It's under doubling because you can only double, not scale by any (arbitrary) ratio, performance has nothing to do with this. The up-/down-scaling algo used after doubling is important for image quality.

So super-xbr is like NNEDI3?? Cant be used to upscale 1280x720 videos to 1920x1080 screens?? So it will double and then downscale. Anyway, what would be the best downscale method after doubling? Im using spline and i get better results doubling it then disable doubling and use luma upscaling alone with jinc :X
xabregas is offline   Reply With Quote
Old 6th July 2015, 20:04   #31554  |  Link
fairchild
Registered User
 
Join Date: Sep 2010
Posts: 321
Quote:
Originally Posted by chros View Post
And 1 more interesting thing: D3D9 exclusive Old path is way-way faster (~40%) then the New path with the same settings (when you switch on several frames in advance)! Is there any advantage (quality wise) to use the New path if everything else seems to work fine with the old one?

Edit: and D3D11 exclusive is the slowest from all them.
Using the same clip, it's the opposite for me, old-path D3D9 is like 19ms rendering times vs 7ms for new-path D3D9. Probably depends on your GPU/drivers and maybe even MadVR settings. The only quality wise advantage I believe is if you plan to use the 10-bit option which requires D3D11 exclusive.

Also D3D11 8-bit is about the same as new-path D3D9 with. (around 7ms)
__________________
MPC-HC/MPC-BE, Lav Filters, MadVR
CPU: AMD Ryzen 5 1600, Video: AMD Radeon RX Vega 56 -> TCL S405 55", Audio: Audio-Technica M50S

Last edited by fairchild; 6th July 2015 at 20:07.
fairchild is offline   Reply With Quote
Old 6th July 2015, 20:49   #31555  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,812
When comparing render times, I think it's important to force the graphics card to run with full clocks.

Maybe this is not possible with GCN Radeons, since Powertune is aggressively lowering GPU clock when there is no full load (which is a good thing, apart from this kind of benchmarks).

I can force my GTX 980 to constantly run with full boost clock when doubling 720p60 with NNEDI3 64.
Under these conditions, the render times of D3D9 and D3D11 are almost identical.
However, I can have huge differences when the GPU isn't running with full clock due to insufficient load.

So I think it's pointless to compare rendering times if the GPU load isn't high enough.
aufkrawall is offline   Reply With Quote
Old 6th July 2015, 20:49   #31556  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
a 24/23p source on a 60 hz display should be faster with old path and that always.

old path is doing more work when the frame rate is not the same as the display rate so it is slower but has more control of what is really happening.

fairchild please check your powerstates for these tests.


using a r9 270
i have 38 ms using spline 3 ar for 720p to 1080p scaling and the reason is lowest powerstate possible even with nnedi3 32 i get lower ms.
huhn is offline   Reply With Quote
Old 6th July 2015, 21:31   #31557  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,812
How big can the impacts of different presentation handling be, compared to Jinc or even Nnedi3?
aufkrawall is offline   Reply With Quote
Old 6th July 2015, 21:51   #31558  |  Link
fairchild
Registered User
 
Join Date: Sep 2010
Posts: 321
Quote:
Originally Posted by huhn View Post
a 24/23p source on a 60 hz display should be faster with old path and that always.

fairchild please check your powerstates for these tests.
I use lav copyback and it forces high powerstates when the clip allows. But I just checked an HEVC clip which doesn't use copyback and it still had same results. 12ms with new path d3d9 and 23ms with old path d3d9.

I am using 14.12 official drivers if that makes a difference, also my settings are pretty pedestrian:

Quote:
ordered dithering with neither option checked
processing-image enhancements: FineSharp - Strength 1.0 (pre-resize manual active for HD)
chroma upscaling: lanczos3 (AR)
image doubling: super-xbr 100 (2x) (active on 960x540 or less)
image upscaling: lanczos3 (AR)
image downscaling: Catmull-Rom (AR+LL)
__________________
MPC-HC/MPC-BE, Lav Filters, MadVR
CPU: AMD Ryzen 5 1600, Video: AMD Radeon RX Vega 56 -> TCL S405 55", Audio: Audio-Technica M50S
fairchild is offline   Reply With Quote
Old 6th July 2015, 22:37   #31559  |  Link
chros
Registered User
 
chros's Avatar
 
Join Date: Mar 2002
Posts: 2,323
Quote:
Originally Posted by fairchild View Post
I use lav copyback and it forces high powerstates when the clip allows. But I just checked an HEVC clip which doesn't use copyback and it still had same results. 12ms with new path d3d9 and 23ms with old path d3d9.

I am using 14.12 official drivers if that makes a difference, also my settings are pretty pedestrian:
Quote:
a 24/23p source on a 60 hz display should be faster with old path and that always.
I'm using 1080p/23.97 content on 1080p/24.01 display (so there's no luma resize) and the framerate is pretty identical. More then that the frame rate switcher is set to these values, and it's working great, and I got the same result with all the different contents:
1080p24, 1080p25, 1080p30, 1080p60

And I wasn't talking about rendering times, but GPU load (though these 2 should depend on each other), using nvidia inspector monitor.
So maybe ATI drivers are behaving completely differently.

Edit: my nvidia driver version is: 347.88 (but I had the same result with the older drivers as well)
__________________
Ryzen 5 2600,Asus Prime b450-Plus,16GB,MSI GTX 1060 Gaming X 6GB(v398.18),Win10 LTSC 1809,MPC-BEx64+LAV+MadVR,Yamaha RX-A870,LG OLED77G2(2160p@23/24/25/29/30/50/59/60Hz) | madvr config

Last edited by chros; 6th July 2015 at 22:39.
chros is offline   Reply With Quote
Old 6th July 2015, 22:55   #31560  |  Link
aufkrawall
Registered User
 
Join Date: Dec 2011
Posts: 1,812
Quote:
Originally Posted by fairchild View Post
I use lav copyback and it forces high powerstates when the clip allows. But I just checked an HEVC clip which doesn't use copyback and it still had same results. 12ms with new path d3d9 and 23ms with old path d3d9.
I can hardly believe these figures. Almost twice as fast, just because of different API?
I'd really rather assume that is somehow a counting/calculation error.
You could check this with madVR settings which almost cost twice as much, e.g. more Needi3 neurons if interop performance is sufficient.
aufkrawall 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 07:23.


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