Optimal pipeline organization for general purpose processors

In this paper we try to conclude what is optimal pipeline organization for general purpose processors. For that purpose we analytically evaluate the performance of eight general purpose processor architectures representing different pipeline organizations in both single processor and multiprocessor setups. The results are interesting. The most efficient pipeline organization for sequential problems is a two-level pipeline with few multiple parallel functional units. The most efficient pipeline organization for parallel problems is a deeply inter-thread superpipelined pipeline in which functional units are chained. Unfortunately these organizations represent the opposite ends of the pipeline organization space. Thus, there exists no single optimal pipeline organization for general purpose processors