Features Overview: Lightspeed Memory Architecture Continued
Higher Order Surfaces
One of the main cause of bottleneck is the strain put on the AGP bus due to the overwhelming amounts of geometry data that needs to transferred between the GPU to the CPU.
Even today's most advanced AGP4X implementation only provides 1.0GB/sec of bandwidth and this is not sufficient for many geometry-rich applications. As an example, it is common for a typical frame in a scene to contain 15 MB of geometry data and for an optimal frame-rate of 60 frames per second, a whopping 900 MB/sec figure is necessitated.
Hence, NVIDIA sought to address this problem and implemented support for higher order surfaces in the GeForce 3.
For rich detailed scenes and objects to be created, a large number of triangles have to be used. This is not entirely efficient especially when sometimes, a majority of them are packed together just to render curved edges with a smooth appearance. As a result, the sheer amount of information that needs to be transferred over the AGP, not to mention the amount of tedious work a designer has to do, is enormous and unnecessary so. A better way to represent complex geometry is with higher order polynomial equations that requires far fewer parameters, translating into more efficient utilization of the limited bandwidth afforded by the AGP bus.
This is a totally important and sensible approach in handling complex geometries within rich 3D scenes. With this method of representation offloading a great deal of data that needs to be transferred over the AGP bus, the congestion is significantly lessened.