Performance of the IPSC/860 Node ArchitectureReport
Intel's iPSC/860 hypercube is the latest in a series of message-passing multicomputers. The performance of individual iPSC/860 computational nodes is the focus of this report; in particular, the performance of basic computational kernels common in scientific computing is examined. Understanding the operation of the iPSC/860 node memory system is key to achieving maximum node performance; from a comprehensive study of the processor-memory interrelationship, guidelines are established for implementing operations in a manner consistent with the processor architecture and memory system performance characteristics. It is demonstrated that the iPSC/860 node architecture exhibits a basic imbalance between processor speed and memory system bandwidth; due to this imbalance, even for highly optimized hand-coded routines the average performance of basic computational kernels can be as much as an order of magnitude below peak processor rate.
Note: Abstract extracted from PDF text
All rights reserved (no additional license for public reuse)
Moyer, Steven. "Performance of the IPSC/860 Node Architecture." University of Virginia Institute for Parallel Computation Tech Report (1991).
University of Virginia, Institute for Parallel Computation