Low-power Instruction Mode Adjustment for 3-D Graphics Processing Unit

碩士 === 國立中山大學 === 資訊工程學系研究所 === 102 === With the rapid development of electronic products, portable electronic products have become the mainstream trend of 3C age. To satisfy high output quality of 3-D image and improve the execution efficiency of 3-D calculations, many portable electronic products...

Full description

Bibliographic Details
Main Authors: Yu-Shan Li, 李鈺珊
Other Authors: Shiann-Rong Kuang
Format: Others
Language:zh-TW
Published: 2014
Online Access:http://ndltd.ncl.edu.tw/handle/fxmt7b
Description
Summary:碩士 === 國立中山大學 === 資訊工程學系研究所 === 102 === With the rapid development of electronic products, portable electronic products have become the mainstream trend of 3C age. To satisfy high output quality of 3-D image and improve the execution efficiency of 3-D calculations, many portable electronic products have adopted 3-D graphic processing units (3-D GPU). However, the large power dissipation of 3-D GPU will cause the portable electronic products, which rely on battery power, to reduce their battery life quickly. Therefore, how to extend the battery life is one of the most important considerations of designing portable electronic products. Until now, the most effective way to extend the battery life is developing the low-power design technique for the 3-D GPU. Owing to the congenital visual limitation of the human eyes, it is difficult for human to quickly differentiate the insignificant distortions of 3-D image. In other words, we can sacrifice a slight accuracy of instructions for reducing the power consumption. To achieve this object, this thesis proposes a low-power instruction mode adjustment system for 3-D GPU. We first replace the single-precision floating-point arithmetic units of 3-D GPU by multi-precision mode arithmetic units, and then develop a fast instruction precision mode adjustment method which can assign the precision modes to multiply-add and special function operations according to the user-defined error constraints. In the case of output image with the acceptable quality loss, we can reduce the power dissipation significantly.