Efficient interconnects for clustered microarchitectures

Abstract

Clustering is an effective microarchitectural technique for reducing the impact of wire delays, the complexity, and the power requirements of microprocessors. In this work, we investigate the design of on-chip interconnection networks for clustered microarchitectures. This new class of interconnects has different demands and characteristics than traditional multiprocessor networks. In a clustered microarchitecture, a low inter-cluster communication latency is essential for high performance. We propose point-to-point interconnects together with an effective latency-aware instruction steering scheme and show that they achieve much better performance than bus-based interconnects. The results show that the connectivity of the network together with latency-aware steering schemes are key for high performance. We also show that these interconnects can be built with simple hardware and achieve a performance close to that of an idealized contention-free model.