Improving the wiki page concerning Anti-Aliasing

[Nocta 3]

Hi guys,

I don't have much knowledge (just starting to learn about it) so I don't really feel comfortable with adding infos on the wiki page but after having talked about it with Mirh, it clear that AA is a big thing and the wiki page here is not up to date neither complete.

 

To quote Mirh:

First of all we should distinguish between post processing AA (which blur image) and proper AA.
Then, FXAA and MLAA should be somewhat penalized in favor of SMAA... and so on.

 

Here you can find some nice resources I collected on the subject if you want to participate and use it in combination with your knowledge to expand the wiki page.

Quite long and exhaustive introduction to AA: http://www.beyond3d.com/content/articles/122/.

A simpler (and older) intro can be found here too: http://www.tomshardware.com/reviews/anti-aliasing-nvidia-geforce-amd-radeon,2868.html.
Interesting topic on AA injection tools: http://www.overclock.net/t/1296721/how-to-anti-aliasing-shader-injection-tools-fxaa-smaa-sweetfx-and-gedosato.

Also a brand new AA techniques seems on his way and impressive: http://www.dsogaming.com/news/epic-games-shows-incredible-new-anti-aliasing-technique-screenshots-video-comparison/

 

Anyone interested and competent? :-D

[Nocta 3]

First post updated with some nice resources. ;)

[primusauloon 7]

That's rather preference based. It seems more apt in an image quality guide regarding subjective AA techniques(post process) so while listing all possible AA present in the game seems fine, ranking them? I'm not too sure about it.

[Nocta 3]

Not necessarily ranking them but giving guidelines for the newcomers could be really useful I think when you see the number of misinformations concerning AA on the web.

[Marioysikax 89]

I would certainly like to see some comparison pictures in AA article both image quality and their impact on frame rate. Also some line between post-process and "real" antialiasing. Also more external links to other sites explaining them. 

 

EDIT: Is there actually any benchmark which would include massive amount of AA technologies which would make making comparison table easier? Games usually simply have post, proper and GPU specific if super lucky.

[Mirh 102]

Unbelievably... even AF filter can be improved :/

...the world is spinning round

[Guest]

Unbelievably... even AF filter can be improved :/

...the world is spinning round

In my experience, unless you're doing QA (Quality Assurance) for a game or will otherwise have issues, forcibly enabling AF to its maximum value globally via your discrete graphics card's Control Panel is perhaps the best tweak ever. All the modern cards I've used, from the GeForce 8600 to the Radeon HD 5670 and beyond, have absolutely no noticeable performance issues from doing so on anything I've thrown at them - multi-monitor configs, downsampled 2K resolutions, you name it. That being said, I wouldn't be surprised if those unnoticeable performance issues became noticeable at 4K - the render costs are exponentially higher there than at 1080P.

[Mirh 102]

Nvidia launched yet another SS technique: Dynamic Super Resuolution

[primusauloon 7]

That's not really a "new" technique, just downsampling except instead of your regular bilinear algo it uses a 13 tap gaussian filter.

[Garrett 211]

Nvidia's resolution feature and Durante's GeDoSaTo are not anti-aliasing methods in the traditional sense; these change the actual resolution seen by the game and so it might have unwanted results (e.g. tiny text/UI on games that don't scale those elements to the resolution).

 

Normal anti-aliasing methods don't have the UI adjustment issue (UI might be blurrier but will still scale as expected).

[Mirh 102]

Normal anti-aliasing methods don't have the UI adjustment issue (UI might be blurrier but will still scale as expected).

Just post processing AA should affect UI sharpness (and textures in general)

[primusauloon 7]

Huh? SSAA is a very valid antialiasing method and still the most effective, the only problem being it's computationally very expensive. Multisampling AA does the same thing as well, except it compromises by only rendering the edges at higher resolution thus saving performance. If by 'traditional' AA you meant MSAA and the like then it's become redundant with more and more usage of deferred rendering(ex:Hitman, Battlefield) and post-AA is the only option we have so far.

As for how much any AA method affects UI elements or subtitles etc, that depends on the developer's implementation if they apply AA before or after the UI layers. Since you mentioned GeDoSaTo, you can actually do that for quite a few games, separate the layers and then apply AA to only the layers you want and ignore the HUD for example(and this is just by using a directx hook with no source code to work with).

[Garrett 211]

I could have worded that better; Nvidia's method works by adding a new virtual resolution to be selected in-game, hence some games having different results (as shown in the article), whereas traditional anti-aliasing methods work with the normal resolution and so don't cause this. GeDoSaTo is a similar concept but avoids the resolution issue by forcing the output in a different way.

[Mirh 102]

As for how much any AA method affects UI elements or subtitles etc, that depends on the developer's implementation if they apply AA before or after the UI layers.

That's exactly why we separate AA into pre processing and post processing methods

[Guest]

That's exactly why we separate AA into pre processing and post processing methods

When using AA methods like FXAA or SMAA, developers also need to split up the post-processing phase between the processed image and things like rendering the HUD on top of it. The reason why is the HUD is usually applied during post-processing (as it typically does not get processed in the same way as polygons, particles, etc.), and if you wait until the image as a whole is completely processed, the HUD will be included in post-processing AA methods - unnecessarily blurring or anti-aliasing elements which were never meant to be affected. Peter "Durante" Thoman has a good explanation of why here.