Analytical methods for the design and optimization of chip-multiprocessor architectures

Abstract

In this paper, we revisit Amdahl's law, expanding it to a generalized form, incorporating energy costs, and exploring its application to chip-multiprocessor architecture optimization. By formulating a system objective function that encompasses the system costs, we can solve for the optimal multiprocessor architecture for a given set of application and architectural parameters. First we use the derived cost function to examine the performance components of two architectures. Second, we apply it to a design example, finding the optimal architecture for a given set of application characteristics.