FSAA Explained

FSAA

Once again we address the question of this term, FSAA. The guys at Beyond3D did a spectacular job with this topic. For an in depth look at this bit of technology take a stroll on down and read their white paper, all mind boggling 20 pages of fun-filled jaggies. Here we will keep it relatively simple.

The information here is pretty much a short summary of what these guys did. 3dfx uses a method called RGSS or JGSS (Rotated Grid Super Sampling, Jittered Grid Super Sampling). NVIDIA uses another method called OGSS (Ordered Grid Super Sampling).

The OG!

OGSS is exactly what it sounds like. Ordered grid means that the image is processed in an ordered fashion. Cut the screen up into nice little blocks and you have an ordered grid. Now we have super-sampling. This means, in really dumb downed terms; that the picture is processed except with a bit more data in it. Mind you, this is all going on within a pixel, thus creating a much more detailed image. The image that is represented at 640x480 is actually processed with as much detail as would be present in something that has, as an arbitrary number, 1.5 times as much detail. So in order for the GeForce2 to display a scene with FSAA at 640x480 it must do the work required for displaying an image at 960x720 and then some. Other stuff like color blending goes on to smooth out the image also. So all in all a considerable amount is going on to create the effect that FSAA delivers.

Voodooss

The 3dfx card does another variation of FSAA called jittered grid super-sampling. JGSS is a derivative of RGSS. RGSS, as opposed to OGSS, takes the image that is going to be represented and processes all data at a slight tilt. Jittered grid has the tilted data set, but it has a randomizing factor thrown in to make it seem more natural. If all the data was rotated at the same angle it wouldn't make too much of a difference in comparison to OGSS. This is because our eyes tend to pick up on patterns relatively easily. The random patterns make sure your eyes don't catch on to what is going on. Following this, all the other color blending and hoo-haa takes place to spruce up the image.

Now what's the deal with 2x and 4x? If you think back a bit, I told you that the screen was split up into tidy little boxes. With 2x sampling assume that two points are used to process all the data within the pixel. With 4x, four points are used. If we have more sub-samples, we have more detail. We also get more points from which to blend and more colors to process. More is better! With more sub-samples other issues come into play, things such as memory bandwidth, memory size, clock speed, all play increasingly greater roles.