Originally posted by: golem
But didn't ATI cause the problem of crapppy AF in the first place with the 9700 series?
ATI was the first to sacrifice IQ in order to offer AF at a minimum framerate hit with R200 (aka 8500/9100). They did this by introducing angle-dependence and by only allowing bilinear filtering between MIP-maps. An argument for this is that 16xAF pushes MIPmap boundaries so far back that bilinear wasn't that big of a deal. The angle-dependency was visible in everyday usage, as Serious Sam and Enemy Territory showed.
Originally posted by: CaiNaM
you remember incorrectly. heh.. r300 started the term, "brilinear".. or has that slipped from memory?
at any rate, ati only filtered certain angles, thereby doing much less work than nv with a resulting performance increase, but at the cost of image quality. at certain angles, it could be considered comparable, but others weren't even being calculated.. and do we have to bring up the whole ati af cheat where they were forcing bilinlear filtering when colored mipmaps were detected?
Actually, IIRC brilinear was first detected on NV cards in UT2k3 (and possibly by Dave in B3D's forums, visible with MIPmap coloring). It was their response to the low performance hit (but also lower IQ) R300 took with AF enabled. ATI later introduced what we dubbed "trylinear" (similar but not exactly like brilinear) in their post-R300 parts, starting with RV350 and extending to the R4x0 series (and possibly R5x0). AFAIK, R300 actually didn't do the trylinear optimization with R300 as it didn't have hardware support for it. ATI did default to trilinear on the first texture layer and bilinear on the remaining ones, though. I can't remember if NV did the same (or if they followed or led). I do know that current ATI drivers have a setting to force trilinear on all layers, which I recently used to great effect in Halo.
Just to be clear, brilinear is not the same as angle dependence. They're separate issues. Bi-/tri-/bri-/try-linear all refer to how cards blend MIP-maps at their boundaries (where a texture sample abuts on that's farther and thus lower-res). AFAIK, angle dependence is simply measuring the angle of a texture relative to the camera and applying more or less AF depending on the angle.
R300 improved on R200 in two ways: it allowed for trilinear, and it was more generous in applying AF (IIRC, it allowed for full AF at 45 degrees in addition to 0 and 90). It was still inferior in terms of quality to NV30, NV25, and NV20 because of its angle dependence, but that was a tradeoff ATI made in favor of speed. R520 appears to finally offer AF approaching GF3/4/FX in quality but closer to 8500/9700 in speed.
I don't remember exactly if NV30 had lower-quality AF than NV2x at launch, or if that came later with brilinear. Either way, it was still technically superior to ATI's (and you can see this in the X850 vs. 7800GT AF tester pics in the link below).
BTW, Cainam, that "cheat" (I agree it was dirty but I don't know if it rises to the level of a cheat, b/c IIRC most ppl were hard-pressed to detect a difference b/w tri and try) was that ATI forced full
trilinear with MIPmap coloring, but resorted to "trylinear" in regular gaming. The simple way around it is to either just play the game or
take some screenshots (er, you can ignore the AF colormaps on the first page

and focus on the actual in-game screenshots on the following pages).