運用Flotherm 配合田口法於伺服器之散熱管理

Thermal Management of a Server by Using Flotherm Package and Taguchi Method

盧峻德(Chun-Te Lu)

指導教授 ： 郭鴻森 ； 李基禎

大同大學/工程學院/機械工程學系所/碩士(2004年)

摘要

隨著電腦核心運算技術的發展，電腦系統散熱的難度也日益增加，如何在專案開發的有限時間內，找出有效且較佳的散熱對策並不容易；因此如何使用模擬軟體配合有效的實驗計畫，使我們能更快的了解影響散熱問題的主要因素，從而分析改善電腦系統內部的散熱機制，使其內部目標元件的操作溫度均能符合其規格要求，為本論文主要的研究方向與目標。

本論文以1U伺服器為例，利用熱流分析軟體Flotherm 4.1並配合田口法做模擬規劃，由田口實驗計畫的結果分析出各散熱機制因素對於目標元件操作溫度的影響，依此規劃出在相關條件限制下的最佳化組合且以模擬預測其結果，最後再配合實測驗證其正確性。此外藉由模擬分析的結果，從中細部分析與探討扁平系統中熱能與流場的傳遞模式，再以目前最高階處理器發熱功率的150W為目標，搭配任何可能改善分析結果的方法，使其晶片表面溫度能低於70℃，試著為新一代的系統散熱問題提出一個有效的解決方案。

經由模擬及實驗分析結果，可以歸納出下列幾點結論：（1）一個良好設計的直交表可以快速找出真正影響目標的因子有哪些；在決定影響因子時要盡可能的將所有主要因素納入，建議以特性要因圖（魚骨圖）的模式將其列出，接著在設計直交表的同時對其做取捨，以避免實驗次數過多影響開發時間。（2）使用模擬軟體建立的模型需要相當仔細，尤其在重要部位的網格要求將直接影響模擬品質的優劣；以本論文中機殼開孔處與內部系統風扇之間的網格設計為例，在未特別加密的情況下，將無法表現出除去第三顆系統風扇且封閉機殼中間開孔時，對CPU的明顯影響。（3）在降低系統阻抗的角度上來看，系統風扇的效能與風罩的搭配對CPU的影響最為直接，其中風罩內是否須加隔離板的問題值得討論；其次為機殼開孔位置與系統風扇數量的取捨，在不超過相關規格的前提下，犧牲次要元件的操作溫度以獲得更多流量的方法也能幫助CPU做有效的降溫。（4）在增加散熱面積上，增加鰭片數量雖然直接，但是對系統阻抗的影響過大，採取將CPU與北橋晶片一起解的方式來增加散熱面積，並且搭配熱管使其在水平傳熱上的效能增加，亦有明顯的成效。

關鍵字

田口法 ； Flotherm ； 伺服器

並列摘要

With the rapid development of central operation technology in computer, to solve the thermal issue is increasingly hard. How to schedule an effective simulation plan to interpret the causes that affect the thermal issue directly, then improve the cooler machinery to keep the operation temperature of main components under their specifications is the main target of this thesis.

In this thesis, use the analysis software Flotherm 4.1 and tie in with the Taguchi method to schedule the simulation plan for discussing the 1U server. To analyze the factor effect on the temperature of main components from the planned simulation result, then scheme the local optimal design and estimate its effect. At last we do an experiment to test and verify the result. Finally, hope to bring up an effective proposal by any possible active solution to solve the thermal problem of next generation. Our target power is over 150W and its’ specification have to less than 70℃.

Making comparison between the final 1U server internal system analysis and the schedule simulation result of Taguchi orthogonal arrays, we have arrived at the following conclusions and recommendation.

1. A full and designate orthogonal arrays can decide the real control factor of quality characteristics. We propose to arrange all main factors with the fish-bone diagram mode and accept or reject them during design the orthogonal arrays.

2. The grid density directly affects the simulation result, especially the area around the main components (like blower, heat sink, CPU and so on). In this case, the result of remove the third system fan could not be manifested, if the density of the grids is not good enough.

3. Consideration was given to reduce the system resistance. The main signal factor is the collocation of fan duct and the blowers, whether to add a separation inside or not is important. The layout of chassis holes and system fans is also a major factor. To keep all components’ specifications, it is possible to raise the airflow of the main portion by excluding some minor one’s temperature.

4. Consideration was given to increase the area of heat dissipation. To add the fin number in the same space directly, it will raise the system resistance too. We recommend solving the thermal issue of CPU and North Bridge chipset together. Simulation comparison revealed that to add the fin length and cooperate with heat pipe would have great improvement in CPU temperature reduction.

並列關鍵字

Flotherm ； Server ； Taguchi Method

參考文獻

7、游裕傑，” 離心式電腦風扇的設計與分析”，國立成功大學機械工程學系碩士論文，2002年6月。

11、S.L.LEE and R.Y.TZONG, ”Artificial pressure for pressure-linked equation”, International journal of heat and mass transfer, vol.35, NO.10, pp.2705-2716, 1992.

1、”Enterprise System Design Conference-Taiwan”, Intel doc#15308, Intel,2004.

2、”RS-Mobile Dothan Processor with 400-MHz FSB Electrical, Mechanical, and Thermal Specifications (EMTS)”, Intel doc#14959, Intel,2003.

3、”Tejas Processor Electrical, Mechanical, and Thermal Specifications (EMTS)”,Intel doc#15663,Intel,2004.