Elite Bastards takes a look at NVIDIA's SLI anti-aliasing

As we've already hinted at in our introduction, the basic theory behind SLI anti-aliasing is a very simple one. In a normal SLI configuration, two video cards are put to work as a team in an attempt to double rendering performance - This is achieved either by each card rendering every other frame in the scene (i.e. GPU one renders the first and third frames, GPU the second and fourth), known as AFR or Alternate Frame Rendering, or by splitting each frame in two and giving a certain portion of the frame to each board, known as SFR or split-frame rendering. The intricacies of these methods are best left for another day, but that's the basic principle.

The concept of SLI anti-aliasing throws this idea out of the window - Rather than having each video card work on separate frames or portions of a frame, instead both boards are put to work rendering each and every frame, as though they were running in a single GPU configuration. The trick to SLI AA is that while each video card renders the scene using the same level of anti-aliasing, the sample patterns used are slightly offset from one another. Then, when both boards have rendered the complete scene for a particular frame, the finished results from both boards are blended together, effectively combining and doubling the number of anti-aliasing samples used and giving a higher level of AA than could normally be achieved by a single board in isolation.

NVIDIA's SLI anti-aliasing investigated