Impact of Computing-in-Memo on the Performance of Processor-and-Menlory Hierarchies
Document Type Article
Abstract
A Hierarchy of Processor-And-Memory can be viewed as an extension of the notion of a memory hierarchy. The extension entails the inclusion of processors with different performance in different levels of the memory hierarchy. Technology trends, applications behavior arid previous research suggest that a rnultiprocessor system organized as a Hierarchy of Processor--4nd-Memory (HPAM) may offer considerable advantages over conventional multiprocessor organizations. This paper analyzes the performance impact of computing-in-memory, for a multi-level HPAM system, as compared to a conventional multiprocessor system with the same cost. The analysis entails using performance models and simulation data. The performance models capture the important differences in the data access time between the two approaches. Underlying the comparative study is the assumption that the cost of a processor is in proportion to the square of its performance, also known as Grosch's law. The 9 benchmarks used in the evaluations belong to the CMI:, Perfect Club and SI3EC95 suites. While the evaluation methodology t,akes into account the heterogeneity of HPAM, the emphasis is placed on modeling the impact of computing-in-memory on the relative performance of the multiprocessors under study. The results indicate that, with rare exceptions, HPAM outperforms conventional multiprocessor designs (of ideni.ica1 cost) by as much a 80% in the benchmarks and the ranges of model parameters considered in the study. In the very few exceptions to this concusion HPAM is never outperformed by more than 20%.