节点文献

计算机系统级热管理方法的研究

Study on the Methods for System-Level Thermal Management of Computers

【作者】 刘明

【导师】 刘静; 周一欣;

【作者基本信息】 中国科学院研究生院(理化技术研究所) , 制冷及低温工程, 2009, 硕士

【摘要】 随着社会信息化进程的加快,计算机技术的发展可谓一日千里。为了获得更快的计算速度和更大的存储容量,芯片的制作工艺不断改进,封装密度不断提高。尽管新技术对单个微器件功耗的降低起到了一定作用,但其已被集成度日益增加所导致的功耗上升完全抵消。因此,热障问题仍制约着当今计算机技术的发展。近年来,为弥补芯片级散热方案的不足,系统级热管理的思想逐步引起高度重视,由此引申出诸多问题亟待解决。本文以获取先进的热管理方案为目标,对应用于桌面计算机、笔记本电脑以及计算机集群的几种新型系统级热管理方法分别进行了理论或实验研究(或两者兼有),在如下几方面取得进展:1.高性能桌面计算机独立热流路径热管理方法的实验研究通过对传统的单一热流路径系统级风冷热管理方法进行数值模拟,分析了该方法在冷却能力和可用性(主要针对噪音)方面的局限。在实验组件的设计中,充分考虑了先前研究中存在的不足,完成了新方案的样机测试,对过程中出现的新问题提出了解决方案。最后通过与传统方案的对比,评估了新方案的先进性。2.笔记本电脑发热元件分离放置的数值模拟分别对笔记本电脑传统的系统级风冷方案和由本实验室前期提出的发热元件分离放置的方案进行了数值模拟。根据模拟的结果,具体分析了后者优于前者的原因所在,并提出了一些实用化解决方案,初步证明了该方案的可行性。3.嵌入机柜式热管理方案的可行性理论分析采用基于exergy损失的评估模型对作者专利中提出的嵌入机柜式系统级热管理方案进行了分析,重点研究了新方案在能量利用率方面的优越性。研究中,首先建立了部分数据中心的简化数值模型并进行了模拟,随后利用模拟结果进一步计算传统方案与新型方案各部分的exergy损失,最后定量分析了新方案的节能潜力,说明了其实用价值。4.嵌入机柜式热管理方案的实验研究组建了两种嵌入机柜式热管理方案的实验样机并采用模拟热源进行了实验测试。对实验过程中出现的新问题进行了理论分析,提出了解决和优化建议。分析了模型与样机间的差别,采用理论分析和数值模拟的方法对实验结果进行了修正,证明了该方案的可行性和实用性。

【Abstract】 With progress of modern informationization, computer technology is witnessing an ever rapid development. In order to achieve a faster computing speed and larger storage capacity, both the chip manufacturing process and packaging density keeps improving. Although some emerging technologies did play a role in reducing the power consumption of a single micro-device, it often encounters trouble caused by the continuously increased power consumption due to dense integration in large-scale integrated circuits. Therefore, the latest computer technology is still restricted by the serious thermal barrier. In recent years, deviating from focusing on chip-level thermal management alone, system-level solution received an ever higher attention which raised a lot of problems for solving. Aiming to establish new techniques for the computer industry, theoretical or (and) experimental researches on several system-level thermal management methods were performed, which could work well in desktop computers, notebook computers and computer cluster, respectively. Results are as follows:1. Experimental study on thermal management method with independent heat flow path used in high-performance desktop computerThrough the numerical simulation on traditional system-level air-cooled thermal management method with single heat flow path, the limitations on cooling capacity and availability (mainly noise) of such method were analyzed. During the process of components design, the flaws in previous research were fully discussed. Experiments on an actual computer were accomplished, and solutions for the raised problems were proposed. Advantages of the new method over traditional ones were also analyzed.2. Numerical simulation on notebook computer with separated heating elementsThe traditional system-level air-cooled method and method by separating heating elements from the main module proposed before by the present lab were simulated. According to simulation results, the reasons why the latter one was better than the former were discussed in detail, some solutions toward practical application were proposed as well, which provided the initial evidence of the feasibility of the new method.3. Theoretical analysis on the feasibility of rack embedded thermal management methodThrough introducing the exergy loss evaluation model, the rack embedded system-level thermal management method proposed by the author in the patent was analyzed, focusing on its superiority in energy utilization. A simplified numerical model for a partial data center was fabricated and simulated. The exergy losses in both the traditional and new methods were calculated based on the simulation results. Finally, the energy saving potential was quantificationally evaluated to illustrate the practical value of the new method.4. Experimental study on rack embedded thermal management methodTwo experimental prototypes of rack embedded thermal management method with simulated heat source were fabricated and tested. Some new problems emerged in the experiment were analyzed theoretically. Solutions and optimization approaches for these problems were proposed. Amendments on experimental results were conducted using theoretical analysis and numerical simulation because of the differences between prototype and actual computer. The results demonstrated the feasibility and practicality of the new method.

节点文献中: 

本文链接的文献网络图示:

本文的引文网络