Originally posted by: Vespasian Is rotated grid supersampling the highest quality implementation of antialiasing?
That's highly debateable, almost any implementation has it's merits and I don't think one can truly pick out any given implementation that is the best at every angle, compatibile with all rendering formats and textures etc.
All in all though, I'd probably give the nod to a truly random RG SSAA implementation as being the best of the 'typical' antialiasing implementations in terms of sheer image quality.
There are certainly more obscure techniques such as prefiltered antialiasing that are almost certainly preferable, but almost all of them have fatal flaws that prevent common usage. For prefiltering that flaw is that it's virtually impossible to implement effectively in a scene in which visibility of any given polygon is not a constant variable... it's great for simple 2D fonts and lines, however it's all but unuseable in consumer level 2D/3D rendering.
Stochastic poisson disk sampling used in combination with SSAA, with 16 samples/pixel might be ideal for games however. It'll reduce aliasing, and the slight increase in noise wouldnt be a big deal for most games.
Of course that's not exactly a viable option as it would effectively kill performance even with the most powerful of graphics cards.
BTW, gamma correct antialiasing is a HIGHLY underrated ability. It's about time that it's being brought to the consumer level in DX9 complaint gaming boards.
It's too bad that consumer level graphics cards have to be so concerned with benchmark performance, otherwise we might see some decent antialiasing implementations.
As is, anything that offers excellent quality usually comes with a performance hit... and 99.9% of reviewers will ignore that benefit in quality and focus purely on the performance hit.
The perfect example of that would be IMHO ATi's R200 > R300 transition, that R200's antialiasing implementation is clearly superior in most respects to that used on the R300. The R300 core is capable of the same implementation but ATi refuses to open it up in the drivers as sites would use it as the "highest quality" mode to benchmark... said benchmarks would put them in an unfavourable light as it's performance would be comparatively low.
Public perception even among the enthusiast community would immediately label it as an inferior implementation regardless of quality purely due to it's somewhat lower performance.
