Skip to Main Content
Future multiprocessor chips will integrate many different units, each tailored to a specific computation. When designing such a system, a chip architect must decide how to distribute the available limited system resources, such as area and power, among all the computational units.In this paper, we introduce MultiAmdahl, an analytical optimization technique for resource sharing among heterogeneous units. MultiAmdahl takes into account the workload, the performance of each computational unit, and the total available resource. The results obtained by MultiAmdahl allow us, for example, to provide a closed-form solution for an optimal asymmetric-offload chip, and to analyze the impact of different design constraints on an optimal chip architecture.
Date of Publication: July-Dec. 2012