多核处理器的加速比研究与热设计
发布时间：2018-09-09 17:55
【摘要】：本世纪初提升主频之路走到了拐点,多核处理器应运而生,逐渐成为研究热点。处理器的重要评价指标是速度和功耗,如何加快速度降低功耗,成为本文关注的重点。暂时条件所限,本工作基于理论计算展开对比研究。 本工作主要目标有二： 一是基于阿姆达尔定律,重新计算多核处理器的加速比,以便分析设计速度更快的多核处理器； 二是基于CMOS电路低功耗设计,研究多核处理器的低功耗设计方法,为研究二维／三维多核处理器热设计做铺垫。 论文的主要内容有四： (1)讨论已有的并行技术研究,由计算时间来衡量处理速度,就产生了加速比的概念； (2)我们评述阿姆达尔定律给出的评价并行计算机加速性能的标准,然后融合兰特法则与阿姆达尔定律,基于第一性原理提出新的加速比描述模型(指数模型)。 (3)讨论对比对称／非对称的多核架构,给出两种架构的时间开销描述式,得到更精确的非对称架构描述式,计算讨论非对称架构优于对称架构的判断条件。 (4)分析电路功耗分布和类型,总结低功耗设计的策略和方法,然后讨论三种多核处理器的热设计方法：热通孔技术；电压岛技术；动态热量管理。 本文主要结果和结论有： (1)经验证,本文提出的新加速比描述模型(指数模型Speedup=mαf)可行。 (2)非对称多核架构在绝大多数条件下的性能都优于对称多核架构。
[Abstract]:At the beginning of this century, the road of improving the main frequency has reached the inflection point, and the multi-core processor has emerged as the times require, and has gradually become a research hotspot. The important evaluation index of processor is speed and power consumption. How to accelerate speed and reduce power consumption has become the focus of this paper. A comparative study is carried out on the basis of theoretical calculation within the limits of temporary conditions. There are two main objectives of this work: one is to recalculate the speedup of multi-core processors based on Amdal's law, so as to analyze and design faster multi-core processors, and the other is to design low-power CMOS circuits. The low-power design method of multi-core processor is studied in order to pave the way for the study of two-dimensional / three-dimensional multi-core processor thermal design. The main contents of this paper are as follows: (1) the concept of speedup is produced when the speed of processing is measured by computing time. (2) We review the criteria given by Amdal's Law to evaluate the acceleration performance of parallel computers, and then merge Rand's Law with Amdal's Law. Based on the first principle, a new speedup description model (). (3) is proposed to discuss the comparison of symmetric and asymmetric multicore architectures. The time cost descriptions of the two architectures are given, and a more accurate description of asymmetric architectures is obtained. (4) analyze the distribution and type of circuit power consumption, summarize the strategies and methods of low power design, and then discuss three thermal design methods of multi-core processor: hot hole technology; Voltage Island Technology; dynamic Heat Management. The main results and conclusions of this paper are as follows: (1) the new speedup description model (exponential model Speedup=m 伪 f) proposed in this paper is feasible; (2) the performance of asymmetric multicore architecture is superior to that of symmetric multicore architecture under most conditions.
【学位授予单位】：苏州大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP332
本文编号：2233135
[Abstract]:At the beginning of this century, the road of improving the main frequency has reached the inflection point, and the multi-core processor has emerged as the times require, and has gradually become a research hotspot. The important evaluation index of processor is speed and power consumption. How to accelerate speed and reduce power consumption has become the focus of this paper. A comparative study is carried out on the basis of theoretical calculation within the limits of temporary conditions. There are two main objectives of this work: one is to recalculate the speedup of multi-core processors based on Amdal's law, so as to analyze and design faster multi-core processors, and the other is to design low-power CMOS circuits. The low-power design method of multi-core processor is studied in order to pave the way for the study of two-dimensional / three-dimensional multi-core processor thermal design. The main contents of this paper are as follows: (1) the concept of speedup is produced when the speed of processing is measured by computing time. (2) We review the criteria given by Amdal's Law to evaluate the acceleration performance of parallel computers, and then merge Rand's Law with Amdal's Law. Based on the first principle, a new speedup description model (). (3) is proposed to discuss the comparison of symmetric and asymmetric multicore architectures. The time cost descriptions of the two architectures are given, and a more accurate description of asymmetric architectures is obtained. (4) analyze the distribution and type of circuit power consumption, summarize the strategies and methods of low power design, and then discuss three thermal design methods of multi-core processor: hot hole technology; Voltage Island Technology; dynamic Heat Management. The main results and conclusions of this paper are as follows: (1) the new speedup description model (exponential model Speedup=m 伪 f) proposed in this paper is feasible; (2) the performance of asymmetric multicore architecture is superior to that of symmetric multicore architecture under most conditions.
【学位授予单位】：苏州大学
【学位级别】：硕士
【学位授予年份】：2012
【分类号】：TP332
【参考文献】
相关期刊论文 前8条
1 张天骐 ,林孝康 ,余翔;SoC系统的低功耗设计[J];单片机与嵌入式系统应用;2004年06期
2 赵国宇;;创建理想的低功耗设计[J];今日电子;2009年02期
3 李文石;姚宗宝;;基于阿姆达尔定律和兰特法则计算多核架构的加速比[J];电子学报;2012年02期
4 王爱峰;李曦;雷霆;周学海;;算法级能耗分析方法研究[J];计算机工程与应用;2006年29期
5 李文石,钱敏,黄秋萍;三维微电子学综述[J];微电子学;2004年03期
6 李文石;二维和三维集成电路的热阻计算[J];微电子学;2005年05期
7 黄国睿;张平;魏广博;;多核处理器的关键技术及其发展趋势[J];计算机工程与设计;2009年10期
8 冯叶;邓倩妮;;非对称多核体系下的阿姆达尔定律性能模型研究[J];微电子学与计算机;2011年08期
相关重要报纸文章 前1条
1 清华大学微处理器与SoC技术研究 中心 汪东升 王海霞 张悠慧 李兆麟;[N];计算机世界;2005年
相关硕士学位论文 前3条
1 陶守春;温度驱动的三维芯片低功耗设计方法研究[D];武汉理工大学;2011年
2 孟一聪;数字集成电路低功耗设计技术的研究及应用[D];清华大学;2005年
3 章承科;基于多核处理器的实时操作系统的扩展[D];电子科技大学;2006年
，本文编号：2233135
本文链接：https://www.wllwen.com/kejilunwen/jisuanjikexuelunwen/2233135.html
