题目/Title：Evaluating the Potential of Graphics Processors for High Performance Embedded Computing

作者/Author：穆帅,王晨曦,刘铭,李东东,朱茂华,陈孝良,谢翔,邓仰东
                        Shuai Mu,Chenxi Wang,Ming Liu,Dongdong Li,Maohua Zhu,Xiaoliang Chen,Xiang Xie,Yangdong Deng

会议/Conference：DATE 2011

地点/Location：Grenoble, France

年份/Issue Date：2011.14-18 March

页码/pages：pp. 1 - 6

摘要/Abstract：
Today’s high performance embedded computingapplications are posing significant challenges for processingthroughout. Traditionally, such applications have been realizedon application specific integrated circuits (ASICs) and/or digitalsignal processors (DSP). However, ASICs’ advantage inperformance and power often could not justify the fastincreasing fabrication cost, while current DSP offers a limitedprocessing throughput that is usually lower than 100GFLOPS.On the other hand, current multi-core processors, especiallygraphics processing units (GPUs), deliver very high computingthroughput, and at the same time maintain high flexibility andprogrammability. It is thus appealing to study the potential ofGPUs for high performance embedded computing. In thiswork, we perform a comprehensive performance evaluation onGPUs with the high performance embedded computing(HPEC) benchmark suite, which consist a broad range ofsignal processing benchmarks with an emphasis on radarprocessing applications. We develop efficient GPUimplementations that could outperform previous results for allthe benchmarks. In addition, a systematic instruction levelanalysis for the GPU implementations is conducted with a GPUmicro-architecture simulator. The results provide key insightson optimizing GPU hardware and software. Meanwhile, wealso compared the performance and power efficiency betweenGPU and DSP with the HPEC benchmarks. The comparisonreveals that the major hurdle for GPU’s applications inembedded computing is its relatively low power efficiency.