The memory wall has become a significant factor in the performance of many scientific applications. Time-to-solution for an application may primarily depend on how efficiently it uses the memory subsystems of the machines on which it runs. MultiMAPS, parameterized by cache hit rates, and specifically designed to fit into a framework for predicting the performance of memory-bound applications is a simple memory benchmark. MultiMAPS measures the achievable bandwidths of memory operations on a machine as a function of cache hit rates.
Download a copy of the MultiMAPS source code distribution: MultiMAPS (revision 3153).
See the README contained within this package for information on how to get started configuring and running MultiMAPS.