The Case for a Multiprocessor on a Die: MoaD

Philip Machanick
Department of Computer Science
University of the Witwatersrand
Private Bag 3
2050 Wits
South Africa
philip@cs.wits.ac.za

Increasingly aggressive pipelining achieves diminishing returns.
Simultaneous Multithreading (SMT) attempts to exploit the fact that
functional units are frequently idle. This paper argues the case for
keeping the processor simple, while looking for parallelism elsewhere,
if it is too hard to find at the instruction level. In particular, it
makes the case for a less aggressive superscalar implementation, and
using the die space instead for a multiprocessor on a die (MoaD), with
the same theoretical peak execution rate.