ATi in for rough times from some reviewers? Aniso on R300.
- Thread starter Hellbinder
- Start date
No one escapes the thunderdome?
I've made mistakes in the past, but I've known full-well about the nature of anisotropic filtering for much longer than three months (In fact, I specifically remember a discussion about it around the time of the GTS wherin I was wondering how the Kyro could support "32-tap anisotropic" when aniso used a variable number of taps...of course the answer was obvious, as DaveB pointed out...).
If you think I made this mistake, then you are projecting somebody else's mistake onto me.
If you think I made this mistake, then you are projecting somebody else's mistake onto me.
Mintmaster
Veteran
Chalnoth said:
Huh? R300 has only 8 texture units - one texture unit per pipeline. In one clock, each pipeline can only do one bilinear sample. I would expect the 9700's anisotropic filtering performance hit to be a bit worse, actually, because of this. However, we won't know for a few weeks...
No, the 9700 can only read one texture per pipeline, but it can do four bilinear samples on that texture.
Remember the original GeForce DDR, for example. It could do either one bilinear-filtered or one trilinear-filtered texture per pipeline. In other words, it could do two bilinear-filtered samples per pipeline. It just had to do them on the same texture.
Anisotropic texture filtering is apparently done today by taking multiple bilinear samples from different positions and combining them. Trilinear is done by taking two bilinear samples and combining them.
Remember the original GeForce DDR, for example. It could do either one bilinear-filtered or one trilinear-filtered texture per pipeline. In other words, it could do two bilinear-filtered samples per pipeline. It just had to do them on the same texture.
Anisotropic texture filtering is apparently done today by taking multiple bilinear samples from different positions and combining them. Trilinear is done by taking two bilinear samples and combining them.
