Jawed
Legend
ralexand said:foaud, why are you adding cost of AA for different sampling rates. You don't do 2X AA then 4X, you do 4X and the performance hit is 5% ie. approximately free. It's not 5% for 2X plus 5% for 4X.
In consumer GPUs AA is normally done in multiple passes (2x per pass), so there is, in effect, a decrease in efficiency dependent on the level of AA - Xenos can only do 2x or 4x AA so there are three coarse steps.
The efficiency losses due to AA in Xenos are caused by:
1. having to copy the entire frame, once rendered, from the back buffer to the front buffer instead of simply flipping between the two as conventional GPUs do. This loss is constant regardless of AA level
2. the (according to the hierarchical Z buffer) visible triangles that straddle tile boundaries will have to be processed multiple times. More tiles means more triangles straddle boundaries.
Jawed