Does the R300 support 1 cycle/pixel trilinear filtering?

[Luminescent]

Why does the performance of the R300 degrade so much when using quality anisotropic filtering? Is it only capable of 8 bilinear filtered pixels per clock? Do you guys believe the R350 will improve the performance of the quality setting (clock-for-clock)?

 

[Dave Baumann]

Its dual cycle trilinear. I quizzed them about this and they cited lack of bandwidth, so if the ratio of bandwidth to pixel pixel pipes is the same on R350 then I doubt there would be changes.

 

[Luminescent]

Are the filtering units actually capable of single-cycle trilinear and only limited by bandwith (to 2 cycles), or did Ati design them with 2-cycle trilinear in mind?

 

[Simon F]

<edit> Decided to delete post of unsubstantiated rumour

 

[Bambers]

http://www.xbitlabs.com/video/radeon9700pro-overclocking/

With 16xAF (quality) in ut2003 there is a noticably bigger improvement overclocking the core than the memory.

 

[Hyp-X]

Are the filtering units actually capable of single-cycle trilinear and only limited by bandwith (to 2 cycles), or did Ati design them with 2-cycle trilinear in mind?

The latter.

You can find the info here

 

[K.I.L.E.R]

http://www.xbitlabs.com/video/radeon9700pro-overclocking/

With 16xAF (quality) in ut2003 there is a noticably bigger improvement overclocking the core than the memory.

I was arguing that a lot too with Smokey from nVnews.
I told everyone I was right.

Anyway, I noticed the fillrate hit with 128tap aniso is very small. So I am thinking that the processing is just done faster.