Performance potential of molecular dynamics simulations on high performance reconfigurable computing systems

The acceleration of molecular dynamics (MD) simulations using high performance reconfigurable computing (HPRC) has been much studied. Given the intense competition from multicore, and from other types of accelerators, there is now a question whether MD on HPRC can be competitive. We concentrate here on the MD kernel computation-determining the force between short-range particle pairs-and examine it in detail to find the performance limits under current technology and methods. We systematically explore the design space of the force pipeline with respect to arithmetic algorithm, arithmetic mode, precision, and various other optimizations. We examine simplifications that are possible if the end-user is willing to trade off simulation quality for performance. And we use the new Altera floating point cores and compiler to further optimize the designs. We find that for the Stratix-III, and for the best (as yet unoptimized) single precision designs, 11 pipelines running at 250 MHz can fit on the FPGA. If a significant fraction of this potential performance can be maintained in a full implementation, then HPRC MD should be highly competitive.