文章摘要
贾凡,章隆兵.基于功耗计数器的处理器功耗实时估算方法[J].高技术通讯(中文),2024,34(1):25~32
基于功耗计数器的处理器功耗实时估算方法
Real-time estimation method of processor power via power counters
  
DOI:10. 3772/ j. issn. 1002-0470. 2024. 01. 003
中文关键词: 功耗估算; 功耗计数器; 功耗管理; 性能事件计数; LASSO回归
英文关键词: power consumption estimation, power consumption counter, power consumption management, performance counter, LASSO regression
基金项目:
作者单位
贾凡 (计算机体系结构国家重点实验室(中国科学院计算技术研究所)北京 100190) (中国科学院计算技术研究所北京 100190) (中国科学院大学北京 100049) 
章隆兵  
摘要点击次数: 808
全文下载次数: 650
中文摘要:
      针对现有的2类处理器功耗实时估算方法的不足,本文开发了一种基于功耗计数器的处理器功耗实时估算方法。该方法结合了基于性能事件计数器和基于电路信号方法的优点,利用功耗计数器记录与处理器功耗密切相关的电路信号的变化次数,能够以较小的观测粒度实时估算处理器的功耗,并且具有较低的硬件开销。本文在龙芯GS364处理器上实现并评估了该方法。结果表明,使用16个功耗计数器可以在512个时钟周期的粒度内实现0.83%的估算误差,与此同时,它的硬件开销只占处理器总面积的0.063%。
英文摘要:
      This paper proposes a real-time estimation method for processor power consumption based on power consumption counters, aiming at the shortcomings of the existing two classes of real-time estimation methods for processor power consumption. The proposed method combines the advantages of performance counter-based and circuit-based methods. Using the power consumption counter to record the number of changes in circuit signals closely related to processor power consumption, the method can estimate the processor’s power consumption in real-time with a small observation granularity, and has lower hardware overhead. This paper implements and evaluates the method on the Loongson GS364 processor. The results show that using 16 power counters can achieve an estimation error of 0.83% within a granularity of 512 clock cycles, and its hardware overhead is only 0.063% of the total processor area.
查看全文   查看/发表评论  下载PDF阅读器
关闭

分享按钮