Computing energy-efficiency in the mobile GPU

In this paper, we explain the energy-efficiency of mobile GPU through the performance and power comparisons of CPU and GPU integrated in the ARM processor-based System-on-Chip. Two kinds of application processors (GPGPU with 72.5GFLOPS computing capability integrated with 1.7GHz dual-core ARMv7A architecture CPUs and GPGPU with 108.7GFLOPS computing capability integrated with 1.8/1.3GHz octa-core ARMv7A architecture CPUs) are used in the experiment. The power and performance analyses of Kishonti Informatics´ CLBenchmark and embedded image processing test suite of Renderscript can provide that the GPU based general-purpose computing ability can deliver 10x higher energy-efficiency in compute-intensive multimedia applications, compared with a CPU solution [1]. The GPU benefits arise from its parallel architecture, which is well suited for compute-intensive multimedia applications. Consequently, a high-performance GPU delivers higher energy-efficiency than any combination of CPU and GPU in most case of compute-intensive workloads.