In this paper we propose a microarchitectural technique called GPU Constant Average Power Processing (GPU-CAPP) that improves the power utilization of power provisioning-limited systems by using provisioned power as much as possible to accelerate computation on parallel workloads. GPU-CAPP uses a ﬂexible, decentralized control to ensure fast response times and the scalability required for increasingly parallel GPU designs. We use GPGPU-Sim and GPUWattch to simulate GPU-CAPP and evaluate its capabilities on a subset of the Rodinia benchmark suite. Overall, GPU-CAPP enables speedup by an average of 26% and 12% over equivalent ﬁxed frequency systems at two power targets.
Kramer Straube, Jason Lowe-Power, Christopher Nitta, Matthew Farrens, and Venkatesh Akella. 2018. Improving Provisioned Power Efﬁciency in HPC Systems with GPU-CAPP,” 2018 IEEE International Conference on High Performance Computing, Data, and Analytics (HiPC), Bengaluru, India, 2018.