Title :
Modeling and Analysis for Performance and Power
Author :
Choi, Jee Whan ; Vuduc, Richard W.
Author_Institution :
Electr. & Comput. Eng., Georgia Inst. of Technol., Atlanta, GA, USA
Abstract :
Accurately modeling application performance for specific architectures allows us to understand and analyze the impact of various architectural features on performance which will ultimately lead to improved performance and better architecture design choices for efficiency and scalability on future systems. However, with the end of Dennard scaling, processors can no longer maintain constant power per unit area as before and consequently power and energy efficiencies has become arguably an even more important factor than performance for future systems. Unfortunately, performance and power/energ efficiencies are not metrics that can be optimized separately, but are intricately dependent factors. Therefore, in order to design efficient systems that can meet both performance and energy/power requirements that are becoming ever more stringent, an acccurate performance-energy model is required so that we can analyze applications on given architectures in terms of both performance and power/energy. With such a model, we can better understand the impact of application and architectural features on performance and power and design applications and systems to maximize their efficiencies.
Keywords :
performance evaluation; Dennard scaling; application performance; architectural features; architecture design; constant power; efficient system; energy efficiency; power efficiency; Analytical models; Computational modeling; Computer architecture; Feature extraction; Graphics processing unit; Lattices; Sparse matrices; energy and power modeling; fast multipole method; lattice quantum chromodynamics; performance modeling; scientific computing; sparse matrix-vector multiply;
Conference_Titel :
Parallel and Distributed Processing Symposium Workshops & PhD Forum (IPDPSW), 2012 IEEE 26th International
Conference_Location :
Shanghai
Print_ISBN :
978-1-4673-0974-5
DOI :
10.1109/IPDPSW.2012.304