Different FSAA implementations.
It's pretty much by design, the GF3/4's 4X MSAA tends to blur textures slightly more then the GF2's SSAA, their similar in respect to how well they remove jaggies and either implementation may have the advantage depending upon the angle and orientation of the aliased textures. The 'twinkling effect' on distinct textures also tends to be more pronounced with the GF3/4 implementation.
It also has issues with transparency and cannot anti-aliased alpha textures which can make some scenes look unusual when only part of the scene is AA'ed.
Of course on the flip side it's quite a bit less memory bandwidth intensive then their prior implementation, and consumes FAR less texture memory.
It was pretty much a trade-off, slightly reduced compatibility and reduced image quality for much better FSAA performance.
ATi went the opposite route with the R8500's FSAA... at least in it's 'quality' mode.
The R8500's pseudo-random sampled super-sampling on the R8500 is a mixed bag.
The "performance" mode is terrible IMHO... comes at a hefty performance hit, and is merely average at removing jaggies and textures are blurred a fair bit.
The quality mode tends to look extremely good though.... it's quite good at removing jaggies, and only blurs textures a small amount.
Unfortunately, it comes at a pretty hefty performance hit, and increases memory usage a lot, as well as being very bandwidth intensive.
In that case they sacrificed performance as it's even slower then traditional SSAA, in favor of image quality.
The opposite was true for their respective anisotropic filtering implementations... there it was ATi sacrificing quality for performance, and nVidia sacrificing performance for quality.
It's all in the nature of their respective implementations, and you seldom get something for free.
There are always trade-offs involved.
Facebook
Twitter