In Situ Power Analysis of General Purpose Graphical Processing Units

Abstract

In this paper, an in situ power analysis profiling over time for general purpose graphics processing units (GPGPU) is presented. Based on this method the power consumption of different modes of operations like data transfer between GPU and host CPU, basic single precision floating point arithmetic operations (addition, subtraction, multiplication) on the multiprocessor units and instructions for shared and global memory access can be measured. There is a factor of 2 difference in power dissipation between various instructions and mode of operations of the GPGPUs. These measurements provide data for an instruction based power estimation of GPU software. It turns out that the power profile over time also gives a good understanding on which section of the program is executed at a certain point in time. The experimental results have been derived from two GPU architectures, namely the GT200 and the GF100 architecture.

Power Measurement Methods for Energy Efficient Applications

“…In [10] the power consumption for several GPU architectures is analyzed, concluding that algorithms can be classified in the two following categories according to its power consumption: data transfer intensive or computationally intensive. However the classification was made over time for GPU executions (transfer and execute kernel) with consumption measured from external power connectors obtaining rough EC values from computing platform, regardless of the communication interface.…”

“…Real-time power consumption values help to develop new power management software techniques, such as power-aware job scheduling. Thereby, identification of power constraints can improve code programming of EEAs [10]. …”

High-level energy consumption model of embedded graphic processors

“…The model can also estimate the energy co component. Shaikh et al [7] profiled the power desktop GPGPU architectures and showed that th of a data transfer instruction consumes less tha kernel instruction. Ma et al [5] chose five m parameters to build their energy model, wh parameters represent the runtime utilizations of stages of the GPU.…”

Power Consumption Prediction Model of General-Purpose Computing GPU with Static Program Slicing

“…GPU 功耗测量是能耗预测的基础.高端 GPU 显卡的供电由两部分组成:主板 PCI-E 总线供电和外接电源. 研究表明,运算能耗主要来自外接电源,PCI-E 总线只提供 10W~15W 的供电,在总体能耗中所占比例甚少,可忽 略 PCI-E 总线供电产生的能耗 [8] .目前,较准确的测量方法是利用电流探头测量显卡外接电源的电流,电流探头 把电流转换为电压信号,再由数字示波器测量电压信号来计算功耗 [7,8] .这种方法具有采样频率高和准确性好的 优点,但是记录采集数据时存在困难. 为了解决实时记录功耗数据的问题,文中设计了功耗采集卡,测量外接电源的电流获得 GPU 能耗.功耗采 集卡工作原理如下:先由电流传感器 ACS713-30T 将电流转换为电压值,采集卡上的微控制器 ATmega168 把电 压模拟信号转化为数字信号;然后,USB 控制器 FTDI FT232RL 芯片将采集数据传输到计算机中保存.该测量方 案不仅采样精度高,而且测量精度高.实验中,应用程序的执行功耗采用如下方式获得:由于在计算过程中功耗 动态变化,因此以程序执行中各采样点的平均功耗作为执行功耗的近似值.设在程序执行过程中各采样点的测 量功耗为{P 1 ,P 2 ,...,P s },则平均采样功耗为该功耗序列的平均值.…”