The Scientific Computing group in the Department of Computer Science
and Engineering has built a 162 processor high-performance Linux
The cluster is for research supported by CEMBA, an NSF-funded
Center for Integrative Graduate Education and Research Traineeship
(IGERT) in Computational Methods for Many-Body Applications.
The cluster was completed and operational in early 2003.
The cluster consists of 81 dual processor compute nodes.
Each compute node has two AMD Athlon MP2200+ processors (CPUs).
Each CPU has a peak floating point performance of 3.639 Gflop/s
(3.639 billion floating point operations per second).
Thus, the machine has a theorectical peak performance of
589.5 Gflop/s. On the Linpack benchmark the machine acheived
a performance of 323.9 Gflop/s on 162 processors (an efficiency
The compute nodes are connected by SCI network cards
manufactured by Dolphin Internet Solutions, Inc.
The cards are two-dimensional D335 PCI cards, and the
compute nodes are connected as a 9 by 9 two-dimensional mesh
The network is extremely high-performance, with point-to-point
latencies and bandwidth of 1.46 µs and 250Mbytes/s respectively.
The computer is programmed using an implementation of MPI
by Scali (ScaMPI) and C and Fortran programs.
Each node is also has two fast ethernet ports which are channel
bonded and used to connect to the server node.
The system was benchmarked on the HP-Linpack benchmark
and acheived a performance of 323.9 Gflop/s which would
rank the machine as the 195-th fastest supercomputer in
the world on the current Top500
list (November 2002).
This performance was obtained on 162 processors in the
solution of an NxN dense linear system Ax = b for N = 93,960.