The geometry calculations across matrices for 3-dimensional euclidean objects are both relatively easy to do directly in hardware and repeated in only slightly varying ways for the various rendering stages.
Given the increasing sizes of framebuffers and textures it makes sense to localize that, although many CPUs now include a GPU equivalent on the same die and many of the same sort of instructions.
A lot of it comes down to how big and independent the working sets are as opposed to strict advantage in computation as well.
Given the increasing sizes of framebuffers and textures it makes sense to localize that, although many CPUs now include a GPU equivalent on the same die and many of the same sort of instructions.
A lot of it comes down to how big and independent the working sets are as opposed to strict advantage in computation as well.