A technique used in
computer graphics, typically
reatime 3D polygon rendering, but sometimes also used with
raytracing. It allows special effects like
motion blur,
depth-of-field and
out-of-focus shots.
The basic idea is very simple. Rather than rendering a single image into a frame buffer, you render several times into the same buffer, each time adding to what is already there. You then average the results.
For example, to do motion blur, you render once with the object, say a spaceship, in its postion at time x, then again over the same buffer at time x+t, then again at x+2t. when you average all three images you get a "blurred" effect.
Similarly, depth of field is caused by having a camera for which the aperture is not a single point. So to get depth-of-field effects, you render the same scene several times, but with the centre of the image each time being slightly displaced, as if it was from some other point other than the perfect centre of the camera.
3dfx calls accumulation buffer a T-buffer for some reason, mostly, it appears, marketing.