An Energy Consumption Model for GPU Computing at Instruction Level

With the development of hardware and software, GPU has been used in Ggeneral-Purpose computation field. The high density of computing resource on chip bring in high performance as well as high power consumption. So the power consumption of GPU has increasingly become one of the most important issue for the development of general computing with GPUs. However, few research focus on estimating the energy consumption of GPU during the computing process. The goal of this study is to build energy consumption model that can predict the energy consumed in the computing phase for various GPUs applications. Our approach is to analyze the PTX code generated by the complier and count the dynamic instruction number that is the challenging problem. The average power can be obtained through this model. And the energy consumption is the product of the average power and the executing time. The experimental results reveal that the average relative error between the prediction model and the measured value is less than 5 percent. It can conclude that the power consumption model from the instruction level can effectively predict the application’ energy consumption.