Super Anti-aliasing - Performance Analysis
I chose Half-Life 2 for testing Super AA mode performance since performance is fairly predictable. It uses AFR for rendering, not Supertiling, so it's more likely to bang up against the CPU limitation wherever possible since the video hardware scales free of shared geometry processing. The nice side effect of that is that Super AA performance should become quite noticeable.What's key to think about before looking at the graph is that, in theory with everything perfect, an AFR-mode game and fast transfers over the PCI Express bus, you can expect 12x AA for the same speed as 6x on a single board, and the same with 4x and 8x. 10x and 14x, the modes with added supersampling, provide their own curiosity.
The minute AFR mode stops being the rendering mode and Super AA is switched on, both cards processing the same geometry to take their respective subsamples, performance goes south. Querying it with ATI upon discovering the performance issue, with the knowledge that sample data has to be pushed over the PCI Express bus which will suck up some performance, solicited a reply along the lines of "yeah, we've got issues just now. A new driver will sort it out so for just now take Super AA as a demonstration of quality rather than potential performance".
We await the new driver for further testing and it seems likely that the driver tested by HEXUS won't make it out into the public domain with the performance issues seen here. That makes it a little redundant to show but HEXUS are committed to evaluating new technology as presented. Something for us to return to soon.
