Why didn't you include the shots without anisotropic filtering for NFS and Q3?
Just a simple graphical depiction.
(The Point)<-------------------------------------------------->(Your interpretation)
Seriously, my fault for not being very clear.
Here's the poop as we know it:
* Alleged drivers that "fix the anisotropic filtering performance issue"
* 29.42s come out with a substantially "lower" anisotropic filteriong performance hit. (which turns out to be an absolutely stunningly small hit between 4x and 8x only. none->2x->4x *performance hit* is exactly the same it has been since the products inception.)
* There is absolutely NO visual difference between 4x and 8x (in Direct3D).
* There is absolutely LITTLE visual difference between 4x and 8x (in OpenGL).
* The 8500 goes up to 16x anisotropic filtering.. and at this maximum level of anisotropic filtering, it's performance hit is *less* than the hit encountered on the GF3/4 at simply 2x.
So, if you figure max anisotropy carries, say a 27% hit on one, and (before) carries a 89% hit on the other.. the simple thing to do to fudge "better" anisotropic filtering performance is to simply cut the process... which apparently has been done on the GF4. Even in Quake3 where 8x does "something"- it can't hardly be miscontrued as 2x the number of texel samples to the scene. And in Direct3D, a smallish (3-8fps) difference in performance that yields absolutely no change at all.
That make sense? I'll be happy to clear up anything else I may not have been to clear on.
