HyperZ is a technology invented by ATI to increase the speed of transferring information from the GPU to video memory .
HyperZ works on a block basis, that is, on the basis of splitting the screen into square fragments. The GPU draws the polygon first in the usual manner, then in the block one, and if the block completely covers the polygon with itself, then it is discarded and excluded from further processing (during rendering this saves up to 20% of the time). Thus, HyperZ saves memory bandwidth , reducing the flow of information passing through it. Consists of three components:
- Hierarchical-Z is a simplified implementation of a hierarchical Z-buffer. The idea of a hierarchical Z-buffer is quite simple: in addition to the Z-buffer itself, a pyramid of Z-buffers of lower resolution is organized. Each of these new buffers has dimensions that are halved horizontally and vertically compared to the previous Z-buffer. In this case, the largest of the four corresponding Z values of the previous buffer with a higher resolution should be written to each Z value from the lower resolution buffer;
- Fast Z-Clear - fast cleaning of Z-buffer. After building and displaying the image on the screen, the information contained in the Z-buffer is no longer relevant and must be erased. The Z-buffer is reset, but not individual values are written, but using blocks, several values are replaced at once;
- Z-Compression - compression of data read and / or written to the Z-buffer without loss.
Content
HyperZ (R100)
The first implementation of the technology.
HyperZ II (R200)
Algorithms for implementing the used functions Hierarchical-Z, Fast Z-Clear, Z-Compression, have been redesigned and improved. Reduced block size (4x4 vs 8x8 on R100). Added the ability to discard 64 pixels per cycle (versus 8 in the R100 ). All this increases the speed by another 20%.
HyperZ III (R300)
Uses a more flexible Z-buffer cache , which is optimized for rendering dynamic shadows in real time.
HyperZ III + (R350)
Improvements affected the Z-cache (a block accelerating work with the Z-buffer ), more efficient compression of Z-data (the maximum value is 24: 1 versus 8: 1 on the R300) and optimization of work with a two-way template buffer. The speed of working with dynamic shadows has been increased, when calculating which the template buffer is actively used.
HyperZ HD (R420)
It is a further development of HyperZ technology, the first version of which was implemented in the first chip of the Radeon family (R100). Postfix HD indicates effective work in high resolutions.