Performance Modeling and Interpretive Simulation of PIM Architectures and Applications (Research Note)

Processing-in-Memory systems that combine processing power and system memory chips present unique algorithmic challenges in the search for optimal system efficiency. This paper presents a tool which allows algorithm designers to quickly understand the performance of their application on a parameterized, highly configurable PIM system model. This tool is not a cycle-accurate simulator, which can take days to run, but a fast and flexible performance estimation tool. Some of the results from our performance analysis of 2-D FFT and biConjugate gradient are shown, and possible ways of using the tool to improve the effectiveness of PIM applications and architectures are given.