TABLE OF CONTENTS The world of general-purpose microprocessors and embedded processor cores has generally settled on power-of-two data-word widths: 8, 16, 32, 64, or 128 bits or the like. Extremely specialized processors or instruction-set extensions to configurable processors sometimes use a less regularly sized data word for specialized instructions, input and output word sizes, and internal registers (even quite irregular and odd sizes such as 13, 17, or 23 bits). However, these specialized, non-power-of-two word sizes are used only where required for very specific algorithms often for data-intensive signal, audio, or video processing where the required precision and range demands the use of a highly-optimized data size to meet system-level cost, power, and performance objectives.
For instance, 24-bit processing seems to be a standard data-word width for high-end audio applications. The 24-bit data word allows for extensive audio processing without audible degradation of the recorded sound and without the overkill of a 32-bit data word. Although the audio samples themselves tend to use only 16 bits (there are 20- and 24-bit audio ADCs and DACs, but they re not yet common), extra bits are needed to provide enough head room to maintain superior sound quality through the intermediate calculations executed by popular digital compression, decompression, and other audio-processing algorithms. However, for general-purpose computing, power-of-two data sizes seem to be the norm.
This was not always the case. The early and middle parts of the computer processor s evolutionary tree contain branches encompassing several...
