NCSA demonstrates results from pilot efforts with Intel on Itanium-based clusters

NCSA has already successfully deployed a 256-processor cluster consisting of Intel Pentium III Xeon processors that ranks 207 on the current Top500 supercomputer sites list (http://www.top500.org/), and is committed to developing a large Itanium-based computer cluster as a solution for its scientific and industrial users. The Alliance also has a 512-processor Linux-based IBM supercluster called LosLobos. That cluster is located at the Albuquerque High Performance Computing Center at the University of New Mexico.

NCSA's demo at The eXHANGE featured two scientific applications that demand top performance: Cactus (http://www.cactuscode.org/) and sPPM (for simplified Piecewise Parabolic Method). Cactus is a multipurpose high-performance toolkit used for computer simulations in a variety of scientific and engineering disciplines. At the eXCHANGE, John Shalf, a researcher at Lawrence Berkeley National Laboratory who works with astrophysicist Ed Seidel and the Cactus team at the Max Planck Institute for Gravitational Physics in Potsdam, Germany, demonstrated wave propagation with Cactus on the Itanium cluster. The Max Planck team has used Cactus to simulate black hole collisions. These kinds of simulations require calculating extremely complex sets of equations, and the results are likely to shed light on fundamental scientific questions, such as the nature of gravitational waves, space and time.

sPPM computes hydrodynamics problems with shocks and is used primarily in astrophysics and defense applications. At the eXCHANGE Paul Woodward, an Alliance researcher at the University of Minnesota, used sPPM to simulate 2D supersonic flow with complex shock interactions. These types of simulations are useful to engineers who need to understand the behaviour of gas flows at supersonic speeds.

"My group has tested the performance of the sPPM code on a wide variety of micro- processors, and this Itanium performance is the best that we have seen to date by a wide margin", said Woodward. "Our codes scale well to thousands of processors, so we expect to be computing at sustained speeds over 1 teraflop (a billion calculations per second) on large Itanium-based computer clusters."

Cactus too has achieved its best performance ever by using the Itanium processor, with a highly optimised version of the code achieving nearly peak speed on the pre-production Itanium-based systems. This performance is almost six times better than performance levels on more conventional supercomputing systems.

Because high-performance Itanium-based computer clusters use the same hardware and software as business and consumer machines, new applications on these clusters should be easy to deploy to a more general audience. Cactus could be used in a variety of business settings, including financial modelling and automotive design. sPPM could be a useful tool for the many engineers that need to consider hydrodynamics problems.