Anisotropic filtering and current consoles.

[BoddoZerg]

Aniso is probably just too expensive (performance wise) for current consoles. If you look at GeForce3/4 level hardware (similar in power to current consoles), Aniso comes with a good sized performance hit. Unless you cheat like the Radeon8500 or GeForceFX, aniso is still pretty costly. I think game developers simply choose to use that performance for other effects, instead of aniso. (Which can be hard to notice in a fast-moving game anyways)

 

[Fafalada]

So then Flipper does aniso sort of like the Radeon 8500 and up then? Where it changes the level of aniso per pixel. So that pixels that need aniso a lot get a high level of aniso and pixels that don't need it as much get a lesser level of aniso.

Pretty much. The actual sampling is different I think, but yeah, idea is the same.

Unless you cheat like the Radeon8500 or GeForceFX, aniso is still pretty costly.

I don't see why you'd have to call it cheating though - it's not like all anisotropic implementations are the same to begin with. Adaptive sampling is an optimization much live various alternative/smarter AA methods, etc.

 

[Tagrineth]

First,

would the person who said GC isnt capable of Anisotropic please never pretend to know anything again.

I said as far as I know.

Second,

I guess PS2 vertex-based could work OK, but it still wouldn't look quite as good as standard, and IIRC that method needs a decent bit of extra texture memory (for Aniso "MIP" maps).

 

[tEd]

afaik nfshs2 uses AF on the xbox .

 

[Fafalada]

I guess PS2 vertex-based could work OK, but it still wouldn't look quite as good as standard, and IIRC that method needs a decent bit of extra texture memory (for Aniso "MIP" maps).

I think some call them rip-maps. And also isn't this just choosing the appropriate rip-map, meaning it's practically free performance wise (at cost of that extra memory)?

Either way, what I talked about is a different approach (described in some SGI paper, I forgot the exact document name). There's no memory overhead, and multipass cost to obtain extra samples is rather similar to GC perpixel implementation.