PDA

View Full Version : Interesting AF implementation analysis


Pages : [1] 2

DSC
12-15-03, 02:30 PM
Well, Muya asked me to start this thread on AF implementations....... :p

So take a look at the link below.

http://www.skenegroup.net/tertsi/aniso/

Lets keep all flamewars etc away, and lets keep the discussion on topic. ;)

The Baron
12-15-03, 02:33 PM
I'd care if it used D3D. OGL has no brilinear from NV, and I don't think it's that interesting of a test. It'll probably just show the theoretical AF implementations (I'm checking it out now).

er. nevermind. I just looked at the R3x0 OGL AF, and I just broke the emergency glass in front of the fire extinguisher. better to be prepared...

saturnotaku
12-15-03, 02:41 PM
Very nicely done. I assume that (dri) means turning on AF in the drivers and (app) is forcing it on through the application itself?

This is pretty much proving what we already know, that NVIDIA cards do a better job of AF at 45-degree and other non-perpendicular angles. What surprises me is that there is some variation between the different settings on the ATI cards.

Were these tested using performance or quality mode in the ATI control panel? Also, are there plans to test using Direct3D?

euan
12-15-03, 04:24 PM
Conclusion:

<noob mode>
ATI - the funny angles are the noes. Close range more detailed.

Nvidia - teh funny angles are nicerer, but the pixal shaders are the noes, and app mode could be better if they didn't mind losing in a couple of benchmarks.
</noob mode>

Rogozhin
12-15-03, 05:27 PM
My main gripe with this article is the lack of info regarding quality-performance settings within the AF tags. If the r3xx was running 8xperformance it was only doing 64tap af instead of the 128tap af when quality is on.

rogo

MikeC
12-15-03, 06:07 PM
Originally posted by saturnotaku
Were these tested using performance or quality mode in the ATI control panel? Also, are there plans to test using Direct3D?

The page has been updated already. Looks like a D3D comparison will be coming.

madshi
12-16-03, 02:30 AM
To me it looks like the OpenGL test program is faulty. When application requested AF looks worse than driver forced AF, most of the time the application has a bug.

The same thing happens with TombRaider. ATI cards have aliasing problems with application requested AF, while NVidia cards have not. An ATI programmer said the reason is that TombRaider requests the anisotropic filtering only for the MIN filter, but not for the MAG filter (whatever that means). ATI's driver/card obeys and the result is texture aliasing. NVidia's driver doesn't accept this strange setting. When the MIN filter is set to anisotropic filtering, it forces anisotropic filtering for the MAG filter, too. This way NVidia works around the application bug.

I assume that the same problem occurs here with the OpenGL test program.

StealthHawk
12-16-03, 02:48 AM
Originally posted by MikeC
The page has been updated already. Looks like a D3D comparison will be coming.

The D3D comparison is up now too :)

AnteP
12-16-03, 02:49 AM
Direct3D, no funny angles:
http://www.skenegroup.net/tertsi/aniso/d3d/aniso13.html

StealthHawk
12-16-03, 03:06 AM
Originally posted by AnteP
Direct3D, no funny angles:
http://www.skenegroup.net/tertsi/aniso/d3d/aniso13.html

Well that's no fun! We already know that the AF is the same without funny angles ;)

The_KELRaTH
12-16-03, 01:24 PM
I would have thought you would also need to test whatever the cards' max AF supported levels are too. It seems rather simplistic to just say that as one card can only use x8AF max therefore all other cards should be subject to the same settings unless of course it's unuseable when implemented for non test apps.

StealthHawk
12-16-03, 08:02 PM
Originally posted by The_KELRaTH
I would have thought you would also need to test whatever the cards' max AF supported levels are too. It seems rather simplistic to just say that as one card can only use x8AF max therefore all other cards should be subject to the same settings unless of course it's unuseable when implemented for non test apps.

ATI cards have the same off angle limitations in 16x AF as they do in 8x AF.

dan2097
12-17-03, 08:41 AM
ATI cards have the same off angle limitations in 16x AF as they do in 8x AF.

But it might look better at the not off angles. This is important as alot of scenary in say an fps wont have problems with off angles (see 4 sided examples)

MUYA
12-17-03, 10:34 AM
From the same geezer

http://www.skenegroup.net/tertsi/aniso/misc/aniso2.html

check it out....a comparision with ATI's AF would be good :D

Jethro
12-17-03, 08:17 PM
well that splains why Nvidia seems so much sharper in T2 with all them rolling hills! neato stuff.

StealthHawk
12-17-03, 08:52 PM
Originally posted by dan2097
But it might look better at the not off angles. This is important as alot of scenary in say an fps wont have problems with off angles (see 4 sided examples)

Yes, it does look better at angles that receive full degree filtering(0, 45, and 90 to be exact). If you look at any review that has image quality comparing 8x AF and 16x AF(and there have been a lot as R300 has been out for a over a year) you can see this :) The R300 launch timeframe is when reviews first started seriously looking at IQ :p

StealthHawk
12-17-03, 08:53 PM
Originally posted by MUYA
From the same geezer

http://www.skenegroup.net/tertsi/aniso/misc/aniso2.html

check it out....a comparision with ATI's AF would be good :D

Does anyone know how to enable 16x AF on NVIDIA cards under OGL?? Or is this perhaps just a typo :confused:

MUYA
12-17-03, 10:27 PM
The perspn who put em them images is an OpenGL developer who said he figured a way to enable 16X AF. and it also states in the Screen captures that it is 16X af. He didn't specifiy how he did except he said that "its has been available in forceware for time time now". I'll try and see how to get to enable this from him :D

frankgom00
12-18-03, 02:14 AM
I own an asus radeon 9800xt and would like to know how to enable negative lod with catalyst drivers

As far as I know ati drivers settings for lod are 0(max quality),+1,+2,+3
no negative lod wich offers a lot better image quality.

Rivatunner doesn't work with my card (catalyst 3.8, 3.9, 3.10, even omega drivers not detected).

So my question is there's any tweak or program that can enable negative lod on ati radeon cards.

Sure it's possible as the article that generated this thread shows.

MUYA
12-18-03, 04:53 AM
OK I asked the guy and he said use 8X AA and 8X AF together to get the erm..16X AF?

MUYA
12-18-03, 05:06 AM
Screenshots of 8X vs 16X in Quake3

http://www.skenegroup.net/tertsi/aniso/games/quake3-aniso1.html

Note the 16X AF shot is with 8X aa enabled along with 8X AF. whereas the 8X af shot is with...8X AF only. Anyone care to analyse that?

Is 8X aa straight msAA or a mixture of ssaa with msaa or something?

StealthHawk
12-18-03, 05:14 AM
Originally posted by MUYA
OK I asked the guy and he said use 8X AA and 8X AF together to get the erm..16X AF?

Well, laugh out loud. Although that is is interesting concept that I planned on looking into. At the time my idea short of fell short and then I decided to scrap it because it would take too much extra effort on my part to devise a new method of presentation. It may be something interesting to revisit though.

frankgom00
12-18-03, 05:15 AM
Please ask this guy how he enabled negative -3.0 lod on an ati radeon card

StealthHawk
12-18-03, 05:16 AM
Originally posted by MUYA
Is 8X aa straight msAA or a mixture of ssaa with msaa or something?

All NVIDIA FSAA modes above 4x a mixture of MSAA and SSAA.

MUYA
12-18-03, 07:24 AM
Ah well, its not 16X AF per say then but, the guy says its similar if not better than ATI's 16X AF implementaion. The mixture os msaaa and ssaa seems to produce real good IQ. Unfortunatly dunno if that is ever playable 8X AA abd 8X AF ;)

He *might* be bothered to do a a comparision shot later with an ATI Card. U know he's busy with work and stuff etc ;)