XPS: FULL: CCA: Cymric: A Flexible Processor-Near-Memory System Architecture

Georgia Tech Research Corporation, Atlanta GA

Investigators

Abstract

Emerging software applications in data analytics, graph analysis, and machine learning must process increasing volumes of data. Such data-intensive applications stress the communication and storage parts of a computer system more than the computation parts. Yet most of today?s systems are ?compute-centric? in the sense that their designs focus on improving the efficiency of computation. Computing near data or processing-near-memory can significantly reduce data communication between memory and processor providing potential for significant improvement in energy-efficiency of next generation computing systems. This project investigates and addresses the challenges facing the design of the next generation of memory-centric processors ? specifically processing-near-memory (PNM) architectures. The project will characterize the energy and time behavior of future applications and use this understanding to assess how to best architect new systems that combine memory and compute in close proximity. Specifically, the activities will address i) programming models, ii) processing-near-memory architectures, and iii) power & thermal managements with the development of cross-cutting data movement, compiler and microarchitectural optimizations. The results of this work can influence the design of future enterprise and high performance computing systems. The major principles, insights, and outcomes of this project will be integrated as modules into mainstream courses in Computer Architecture. The investigators will also continue their participation in Institute outreach programs that have the goal of promoting participation of undergraduate, minority, and female students in engineering research and higher education.