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 20th October 2016, 23:01   #39781  |  Link
Shiandow
Registered User
 
Join Date: Dec 2013
Posts: 753
Quote:
Originally Posted by sauma144 View Post
Is it SSimSuperRes, its successor or a totally new algo?
What's the performance difference between NNEDI3 (64n) and your algo?
For a large part it's completely new, but it's somewhat similar to the Bilateral chroma scaler. On my GTX 960 it takes about 2~3 times as long as bicubic, or about an additional 0.75ms. Still need to optimise it some more though. At any rate it should be significantly less performance intensive than NNEDI3.
Shiandow is offline   Reply With Quote
Old 20th October 2016, 23:06   #39782  |  Link
CarlosCaco
Registered User
 
Join Date: Oct 2015
Location: Brasil, SP, São Paulo
Posts: 154
Quote:
Originally Posted by Shiandow View Post
For a large part it's completely new, but it's somewhat similar to the Bilateral chroma scaler. On my GTX 960 it takes about 2~3 times as long as bicubic, or about an additional 0.75ms. Still need to optimise it some more though. At any rate it should be significantly less performance intensive than NNEDI3.
Is this the new algo madshi talked about?
__________________
Desktop, i5 2500, 8GB, N570 GTX TF III PE/OC
Asus X555LF, i7-5500U, 6GB Ram, Nvidia 930m/HD 5500
Windows 8.1 Pro x64
CarlosCaco is offline   Reply With Quote
Old 20th October 2016, 23:40   #39783  |  Link
sauma144
Registered User
 
Join Date: Sep 2016
Posts: 89
Quote:
Originally Posted by Shiandow View Post
For a large part it's completely new, but it's somewhat similar to the Bilateral chroma scaler. On my GTX 960 it takes about 2~3 times as long as bicubic, or about an additional 0.75ms. Still need to optimise it some more though. At any rate it should be significantly less performance intensive than NNEDI3.
So it is a chroma only scaler?
I hope I am wrong but it seems that the MPDN development is in standby.
Your previous algorithms/shaders were initially made for MPDN.
Do you still create them for MPDN?
Is there an hidden option in madVR to use external shaders?
sauma144 is offline   Reply With Quote
Old 20th October 2016, 23:44   #39784  |  Link
Ver Greeneyes
Registered User
 
Join Date: May 2012
Posts: 447
Quote:
Originally Posted by Shiandow View Post
True, although softer images almost automatically look less aliased.
Yeah, and I do think the sharpening looks good (though I can't judge whether it's accurate to the original). It's mostly the topmost fold of cloth on the sleeves of both men that looks a bit jagged where NNEDI3 makes them look like smooth gradients. If you could fix that I think it would improve upon NNEDI3 in that image, assuming the sharpness is 'correct' (for some sense of the word).
Ver Greeneyes is offline   Reply With Quote
Old 21st October 2016, 00:32   #39785  |  Link
Shiandow
Registered User
 
Join Date: Dec 2013
Posts: 753
Quote:
Originally Posted by CarlosCaco View Post
Is this the new algo madshi talked about?
Probably not.

Quote:
Originally Posted by sauma144 View Post
So it is a chroma only scaler?
[...]
Do you still create them for MPDN?
Is there an hidden option in madVR to use external shaders?
Yeah, it's meant just for the chroma. It's basically another attempt at trying to recover chroma information from luma. So far it seems to be more successful than other methods I tried, although very small details can be a bit problematic. To avoid going too far off topic I won't go into depth on the current state of MPDN but I still use it to try out new algorithms. No hidden options for injecting shaders in MadVR, unfortunately.
Shiandow is offline   Reply With Quote
Old 21st October 2016, 09:05   #39786  |  Link
adhara
Registered User
 
Join Date: Jan 2013
Posts: 41
Hi all,

This is about RGB output levels with a 10 bits display (so dithering 10bits and more activated in MadVr).
I know this is not a new question but I would like a clear explanation as we can read everything and its opposite arround this subject.

Basically there are 3 options:

Setup 1 :
GPU outputs 0-255
MadVr outputs 0-255
Display expects 0-255

Setup 2:
GPU outputs 16-235
MadVr outputs 16-235
Display expects 16-235

Setup 3:
GPU outputs 0-255
MadVr outputs 16-235
Display expects 16-235

Problems:
-There is no way on my UHD (10 bits) TV to select limited or full range.
- Graphic Card GTX 1070: When output is set to 2160p(4k) - RGB Full range, bitdepth is limited to 8 bits / To have 10/12 bits outptut on the GPU, I have to lower outptu level to limited range (16-235).

So what is the best setting for my own use (4k) ?

Madvr (0-255 / 10 bits) --> GPU (0-255 / 8 bits) --> Display

OR

Madvr (16-235 / 10 bits) --> GPU (16-235 / 10 bits) --> Display

OR

Madvr (0-255 / 10 bits) --> GPU (16-235 / 10 bits) --> Display

OR (other ?)

Thanks
adhara is offline   Reply With Quote
Old 21st October 2016, 09:56   #39787  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by Shiandow View Post
Seems I may need to hurry along my latest experiment a bit.
I think it looks too sharp. NNEDI3 isn't the one to beat anyway, it's super-xbr AR. Considering you seem to like working with bilateral for chroma I can only assume it has benefits but I haven't seen anything that makes me consider it yet. I've wondered if it would be an interesting idea to use this or something similar as a second pass for chroma on top of super-xbr and averaging the result.
ryrynz is offline   Reply With Quote
Old 21st October 2016, 11:46   #39788  |  Link
Shiandow
Registered User
 
Join Date: Dec 2013
Posts: 753
Quote:
Originally Posted by ryrynz View Post
I think it looks too sharp. NNEDI3 isn't the one to beat anyway, it's super-xbr AR. Considering you seem to like working with bilateral for chroma I can only assume it has benefits but I haven't seen anything that makes me consider it yet. I've wondered if it would be an interesting idea to use this or something similar as a second pass for chroma on top of super-xbr and averaging the result.
Blending the images would probably get you the worst of both worlds. Anyway, here is a rather extreme example that shows why I like 'bilateral' methods. Sure in most cases the difference is more subtle, but it's the cases where it's not that matter. The difference also becomes more noticeable if you use upscaling. By the way, if I recall correctly Chroma-SuperXbr also uses some luma information, just not as extensively.
Shiandow is offline   Reply With Quote
Old 21st October 2016, 11:51   #39789  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by Shiandow View Post
here is a rather extreme example that shows why I like 'bilateral' methods.
Those jaggies tho..
ryrynz is offline   Reply With Quote
Old 21st October 2016, 12:00   #39790  |  Link
Shiandow
Registered User
 
Join Date: Dec 2013
Posts: 753
Quote:
Originally Posted by ryrynz View Post
Those jaggies tho..
Yeah. most are the result of not using the full luma information. As far as I can tell Superxbr isn't doing much better. I managed to beef things up a bit and get this which is somewhat better, but I need to check if I can raise the strength that much without making other images worse.
Shiandow is offline   Reply With Quote
Old 21st October 2016, 16:23   #39791  |  Link
Stereodude
Registered User
 
Join Date: Dec 2002
Location: Region 0
Posts: 1,436
Quote:
Originally Posted by Asmodian View Post
FLOPS, and I agree that it seems to indicate madVR performance quite accurately.
I don't think you can use FLOPS to compare across hardware generations.
Stereodude is offline   Reply With Quote
Old 21st October 2016, 17:39   #39792  |  Link
Asmodian
Registered User
 
Join Date: Feb 2002
Location: San Jose, California
Posts: 4,406
Quote:
Originally Posted by Stereodude View Post
I don't think you can use FLOPS to compare across hardware generations.
I don't know for sure but it seems like you can, it is not super precise but significantly higher FLOPS gives higher performance. FLOPS seems to be a better indicator for madVR performance than for most other types of loads.

Like we were discussing earlier, FLOPS doesn't give a 1:1 performance boost, but for NNEDI3 the card with more FLOPS runs NNEDI3 faster.
__________________
madVR options explained
Asmodian is offline   Reply With Quote
Old 21st October 2016, 17:41   #39793  |  Link
Georgel
Visual Novel Dev.
 
Georgel's Avatar
 
Join Date: Nov 2015
Location: Bucharest
Posts: 200
Quote:
Originally Posted by ryrynz View Post
I can't see any difference either...

Quote:
Originally Posted by Shiandow View Post
Yeah. most are the result of not using the full luma information. As far as I can tell Superxbr isn't doing much better. I managed to beef things up a bit and get this which is somewhat better, but I need to check if I can raise the strength that much without making other images worse.
SuperXBR has a bit of jagged edges, but still, those results do show a bit too...

At that point NNEDI, bilateral or reconstruction or even Jinc look better. At least if you're alergic to jagged edges.


By the way, anyone knows if there is a big difference in image downscaling between SSIM 2D and 1D, or rather what is the difference?
Georgel is offline   Reply With Quote
Old 21st October 2016, 18:32   #39794  |  Link
leeperry
Kid for Today
 
Join Date: Aug 2004
Posts: 3,477
Quote:
Originally Posted by Georgel View Post
SSIM 2D and 1D (..) what is the difference?
jinc/bicubic
leeperry is offline   Reply With Quote
Old 21st October 2016, 20:31   #39795  |  Link
har3inger
Registered User
 
Join Date: Feb 2014
Posts: 139
I'm a big fan of bilateral with some superchromares, so I'm looking forward to your new scaler, shiandow. Generally, every other non-reconstruction chroma scaler will look thick and fuzzy due having half the resolution, but bilateral cleans things up nicely, as it basically works with built in line thinning. Artifacts are cleaned up a bit with SR at the cost of chroma cleanliness.
har3inger is offline   Reply With Quote
Old 21st October 2016, 22:57   #39796  |  Link
Stereodude
Registered User
 
Join Date: Dec 2002
Location: Region 0
Posts: 1,436
Quote:
Originally Posted by Asmodian View Post
I don't know for sure but it seems like you can, it is not super precise but significantly higher FLOPS gives higher performance. FLOPS seems to be a better indicator for madVR performance than for most other types of loads.

Like we were discussing earlier, FLOPS doesn't give a 1:1 performance boost, but for NNEDI3 the card with more FLOPS runs NNEDI3 faster.
The data seems to suggest that's very much not true. The GTX 1060 outperforms the RX480 (just barely) in the madVR data posted despite having 33% less FLOPS. The same data shows the GTX 1060 outperforming the GTX 970 by >20% despite having only a 10% FLOPS advantage.

I wish we had data comparing an R9 card with basically the same FLOPS as the RX 480. I'm fairly confident it would show the RX 480 inferior despite comparable FLOPS.
Stereodude is offline   Reply With Quote
Old 22nd October 2016, 03:56   #39797  |  Link
seiyafan
Registered User
 
Join Date: Feb 2014
Posts: 162
Two years I had a 270x and GTX 770 and they were about the same in madVR even though the 770 had more FLOPS. So now it seems the opposite is true.
seiyafan is offline   Reply With Quote
Old 22nd October 2016, 09:16   #39798  |  Link
huhn
Registered User
 
Join Date: Oct 2012
Posts: 7,903
the 1060 has less gflops at the stock boost clock but every 1060 clocks way hjigher than the "max" boost clock that makes normal Gflop calculation pretty much worthless.

than we have openCL efficiency in the past AMD was way way better at this than nvidia even with copyback bug.
huhn is offline   Reply With Quote
Old 22nd October 2016, 10:18   #39799  |  Link
madshi
Registered Developer
 
Join Date: Sep 2006
Posts: 9,140
Sorry guys, my new algo needs some more tweaking time. Maybe 1-2 weeks more, I don't know. Quality is constantly improving, though. Performance seems ok, but will require a decent GPU.

Will be back to answer all open questions/comments when the new algo is completed.
madshi is offline   Reply With Quote
Old 22nd October 2016, 11:49   #39800  |  Link
ryrynz
Registered User
 
ryrynz's Avatar
 
Join Date: Mar 2009
Posts: 3,646
Quote:
Originally Posted by madshi View Post
Will be back to answer all open questions/comments when the new algo is completed.
At least this.. upscaler or post processing?
ryrynz 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 22:43.


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