On the Radeon R100-based cores, Radeon DDR through 7500, where HyperZ debuted, ATI claimed a 20% improvement in overall rendering efficiency. They stated that with HyperZ, Radeon could be said to offer 1.5 gigatexels per second fillrate performance instead of the card's apparent theoretical rate of 1.2 gigatexels. In testing it was shown that HyperZ did indeed offer a tangible performance improvement that allowed the less endowed Radeon to keep up with the less efficient GeForce 2 GTS.[5]
Functionality
HyperZ consists of three mechanisms:
Z compression
The Z-buffer is stored in a lossless compressed format to minimize the Z-Buffer bandwidth as Z read or writes are taking place. The compression scheme ATI used on Radeon 8500 operated 20% more effectively than on the original Radeon and Radeon 7500.
Fast Z clear
Rather than writing zeros throughout the entire Z-buffer, and thus using the bandwidth of another Z-Buffer write, a Fast Z Clear technique is used that can tag entire blocks of the Z-Buffer as cleared, such that only each of these blocks need be tagged as cleared. On Radeon 8500, ATI claimed that this process could clear the Z-Buffer up to approximately 64 times faster than that of a card without fast Z clear.
Hierarchical Z-buffer
This feature allows for the pixel being rendered to be checked against the z-buffer before the pixel actually arrives in the rendering pipelines. This allows useless pixels to be thrown out early (early Z reject), before the Radeon has to render them.
Versions of HyperZ
With each new microarchitecture, ATI has revised and improved the technology.
