A new era in supercomputing has arrived at UB, one of just three sites in the world selected by SGI to beta-test Intel's next-generation chip, the new Itanium processor. The other sites are the Ohio Supercomputing Center in Columbus and the University of Manchester in the United Kingdom.

"This is the next wave of supercomputing," said Russ Miller, professor of computer science and engineering, and director of the Center for Computational Research (CCR).

CCR is conducting the tests in close cooperation with SGI, which has integrated the Itanium processors into a PC-based system; CCR scientists are networking the PCs together to create a small cluster-based system.

"The Itanium, Intel's next-generation chip, represents not just a dramatically more powerful chip eventually capable of performing computations almost an order of magnitude faster than the current Pentium III," said Miller, "but it also is the first time that mass-marketed chips, known as commodity processors, will be used in supercomputers."

The Itanium is expected ultimately to displace the current Pentium processor in a wide range of computing applications.

"Eventually, this is going to drive the costs of supercomputing way down," said Miller. "Unlike previous supercomputers that were expensive due to proprietary, special-purpose hardware and software, these systems will rely on commodity processors, commodity software and new generations of hardware and software to connect these processors."

The choice of UB as one of only three beta-sites worldwide reflects the breadth and depth of CCR's research focus, particularly in modern structural biology, said Miller.

It also reflects the center's extensive experience with cluster computing, where scientists cluster together anywhere from a few to many commodity-based computers in order to achieve-or surpass-the power of a single supercomputer at significant cost savings. CCR scientists have created clusters based on Alpha processors, Pentium processors and Sparc processors, among others, connected with a variety of networks.

A key test for the new Itanium technology at CCR will be how it performs with SnB, the protein-structure software of choice used by more than 500 drug-design-and-research labs that was developed by scientists at CCR and Hauptman-Woodward Medical Research Institute.

SnB, based on the "Shake-and-Bake" algorithm developed by Nobel Laureate Herbert Hauptman, has become a critical tool for determining atomic or near-atomic structures for biomedically important molecules. It is expected to gain even wider use now that the human genome has been sequenced and scientists have millions of gene structures to solve.

"SnB is a computationally intensive program that has been tailored to run on a wide variety of platforms," said Miller. "Therefore, we are able to use the program to evaluate the relative power of many different computer platforms."

Miller said CCR is "stressing" the Itanium-based system in order to evaluate the performance of university-developed codes and to evaluate SGI's Fortran and C compilers, which take codes written by computer scientists and translate them into languages machines can understand.

"An Intel Itanium cluster provides amazing potential in terms of providing supercomputing capabilities at a significantly reduced price," said Miller. "It will provide small laboratories with the option of obtaining cost-effective computing solutions to problems that may be scaled to large systems in a relatively seamless fashion. Building these systems on top of Linux, an open-source operating system, allows scientists, engineers and administrators to tune the systems for their specific needs.”

According to Miller, the results of CCR’s tests with the prerelease Itanium chips bode well for their performance in supercomputers.

“In evaluating the SnB system on our Beta SGI Itanium cluster, it appears that the prerelease IA-64 (Itanium) processors already are competitive with high-end processors,” said Miller.

“We anticipate that for SnB, these processors will outperform many of the high-end processors when finally released.

“Further, if these processors are available at a reduced cost compared to high-end processors, then the ability to deliver quick turnaround time for computationally intensive programs like SnB at low cost will be enhanced greatly,” he said. “This is certainly good news for the genomics community.”