Hardware Geometry Setup
The hardware triangle setup engine is what makes 3D accelerators… accelerate 3D. The first generation of 3D cards relied on the CPU to process polygons before sending them off to be rendered. A full hardware triangle setup engine bypasses the CPU and does the triangle setup on the video card. This frees up a considerable amount of CPU processing power.
Take a load off
If the triangle setup engine takes the load off the CPU, why don't you get the same 3D frame rates on my p200mmx as your p3 500? Well, the CPU still has to perform transform and lighting tasks, but future 3D accelerators are also going to move T&L onto the video card. Nvidia's upcoming NV10 chip is going to feature onboard T&L, and you can expect that any other competitive next-gen products (3dfx and s3) will as well, on top of other performance-enhancing features.
As 3D accelerators become less CPU dependent, we'll begin to see cards that provide similar framerates regardless of CPU speed. The framerates on a p3 450 will be the same as those on a p3 500. Once lighting and transform functions are integrated onto the video silicon, the task of the CPU will be greatly reduced, to the point where the all-important floating-point power of a CPU will no longer matter (at least to such a degree).
As a quick example, Nvidia is estimating that NV10 will perform at four times the speed of TNT2. Barring any MMX/SSE/3DNow! hype to the same degree, real-world examples of hardware geometry acceleration lends credence to these claims. Of course, it also means that it will matter less whether you have a P2 400 or a P3 666. This is good news to most folks, and it also means that the term "CPU-limited" won't be used nearly as much.
Another benefit to hardware geometry setup is that the freed-up CPU can be used for so many more pressing tasks. In essence, it would be akin to running a dedicated server. Programmers would be able to spend much more time working on realistic physics, AI, and other computationally intensive algorithms.