Anisotropic Filtering & AA

[40sTheme]

I've always wondered how exactly this works. What exactly does it do? How does it filter compared to trilinear filtering? I've tried looking at some explanations, but they really don't help too much.
Also, with Anti Aliasing, what is the difference between multi sample, super sample, and transparency AA?
And for both, what do the x4, x6, etc mean? I think it means the number of samples.
I understand what they do, but I want to know how and why they do it in more depth.

 

[m21s]

Should answer most of your questions.


http://en.wikipedia.org/wiki/Anti-aliasing

http://en.wikipedia.org/wiki/Anisotropic_Filtering

 

[40sTheme]

Wow, that's alot of information to take in. =0
I'm gonna put some of those anti-aliasing formulas into Java with a random assignment to the variables and see the outcomes based on each method.
I betcha it'll be cool. =)

 

[m21s]

:thumbsup:

 

[BFG10K]

Some fast answers:

Anisotropic filering: employs non-square sampling patterns in the direction of the stretch which keeps textures sharper at angles. It makes the scene sharper than bilinear/trilinear, especially at steep angles.

Anti-aliasing: removes jagged edges by sampling multiple data elements (the elements depend on the AA method). The number used is usually the number of multi-samples being taken.

Multi-sampling: uses Z tests to anti-alias polygon edges and hence samples geometry only.

Super-sampling: samples everything including texels.

Transparency: samples alpha (transparent) textures only.

 

[40sTheme]

Hmm, so super sampling would definitely use more video memory than multi sampling. Transparency is probably the best since it samples the actual base structure of the texture though, right? Or am I confusing myself?

 

[Gamingphreek]

Originally posted by: 40sTheme
Hmm, so super sampling would definitely use more video memory than multi sampling. Transparency is probably the best since it samples the actual base structure of the texture though, right? Or am I confusing myself?

Well you are right on your first point. Super Sampling does use significantly more memory than multi sampling.

However transparency is not the best. It is more function specific. If you remember what BFG said, "It only samples alpha textures". Therefore, the edges of non alpha textures would not be improved.

-Kevin

 

[tuteja1986]

lalalalalalal 🙁

Wrong thread

 

[Gamingphreek]

Edited

 

[BFG10K]

Hmm, so super sampling would definitely use more video memory than multi sampling.

Yes, it uses more bandwidth because multi-sampling always does one texture lookup regardless of the amount of multi-samples being taken, unlike super-sampling.

Transparency is probably the best since it samples the actual base structure of the texture though, right?

Transparency AA is the best in its class for exactly that: transparent textures. However it won't work for anything else.

The best AA method will mix rotated grid multi-sampling with full screen super-sampling and stack with transparency AA as well. nVidia offers this mode on single cards while ATi requires Crossfire to do it.