ARCHIVED: What is a GRAPE, and what GRAPEs are available at IU?

A GRAPE (GRAvity PipE) is a special-purpose computer designed to tackle the N-body problem where a collection of "particles" interact with each other through some force.

The GRAPE-6 system is designed to calculate the 1/r² force of gravity. At Indiana University, a GRAPE-6 board with a peak processing capacity of one teraflop is installed in two Gentoo v1.6-based 2.2GHz AMD Opteron systems named g6a-g6b.uits.iu.edu. With this system, the gravitational forces on one million virtual particles can be calculated.

The Molecular Dynamics GRAPE (MDGRAPE) is another type of GRAPE processor. Whereas gravity is only an attractive 1/r² force, the electrical force can be either an attractive or a repulsive 1/r² force. In fact, the MDGRAPE can handle any arbitrary force law or potential that can be expressed mathematically between two boundary points. Two MDGRAPE-2 cards, each capable of a peak theoretical processing capacity of 64 gigaflops, are installed in four RedHat v7.3-based 700MHz Pentium III systems named mdgrape01-mdgrape04.uits.iu.edu. IU acquired two additional MDGRAPE-2 cards in July 2004. At that time, both the existing cards and the new cards were equipped with booster cards, which essentially doubled the performance of each card to 128 gigaflops. IU's four MDGRAPE-2 cards will be made available for use as a TeraGrid computing resource later this year.

The function call to the GRAPEs also returns a list of neighbors. This information can be useful for other types of codes. Two examples are Smooth Particle Hydrodynamics (SPH) and "Tree" codes. These codes are not interested in calculating the interaction of one particle with all the others, but only the interactions with their nearest neighbors. By using the GRAPEs, neighbors can be determined with one call to the GRAPE hardware. The system ignores force or energy results and uses only the list of neighbors in further calculations on the host computer.

For more detailed information about the GRAPEs, see:

This document was developed with support from the National Science Foundation (NSF) under Grant No. 0503697 to the University of Chicago and subcontracted to Indiana University. Additional support was provided by IU through its participation in the TeraGrid, which is supported by the NSF under Grants No. 0833618, SCI451237, SCI535258, and SCI504075. Any opinions, findings, and conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the NSF.

Comments/Questions/Corrections