UA Makes the TOP500 Supercomputer List

Lori Stiles
Nov. 20, 2000


The University of Arizona's supercomputer, known as "Super," has made its first appearance on a prestigious ranking of the most powerful computers on the planet.

The 16th edition of the TOP500 list, which is sponsored by the University of Mannheim (Germany) and the University of Tennessee, was released earlier this month in Dallas at SC2000, a conference for professionals involved in high-performance networking and computing.

Of the top 500 computers, 233 are located within the United States. The university's Super ranks 389 on the world list -- and 16th among the 20 supercomputers listed at U.S. universities. A total 86 academic institutions world-wide made the new TOP500 list, which has been updated twice a year since June 1993.

"Being among only 20 universities rated as top supercomputer sites in the country is an important achievement for the UA," said Vice President for Research Richard Powell. "Our supercomputer provides a critical facility for our students and faculty members involved in high performance computing."

Super's inclusion on the list is the result of two major system upgrades within the past year. The upgrades more than doubled the number of processors in the system, as well as Super's computing capabilities.

A 5,000-pound, shared-memory machine with 104 processors and 48 gigabytes of memory in an assembly of modular racks six feet high, four feet deep, and 20 feet long, Super can run several large jobs or numerous smaller jobs at the same time. Super is housed and maintained by the UA's Center for Computing and Information Technology (CCIT).

Rankings on the list are determined by running benchmarking software--a set of standardized tests to measure system performance--on the supercomputer.

In this round, the IBM at Lawrence Livermore National Laboratory took top honors with a speed of 4938 gigaflops. The UA machine, a Silicon Graphics Origin 2800 (SGI), clocked in at 66.6 gigaflops. One gigaflops is a thousand million floating point operations per second. (The "s" in gigaflops stands for second, so this word always ends in "s".)

Any UA faculty member who teaches can get user privileges and can sponsor privileges for other UA staff and students who are doing research.

Primary users are high-performance computing groups from engineering and scientific disciplines. The most common job: running simulations on electronic structure, weather forecasting and weather-pattern studies, computational fluid dynamics, solid- state physics, and astrophysics, among others.

"I think it's fair to say this is a historic moment because this is the first time we can honestly claim we are joining the big league in supercomputing," said Herman F. Fasel, professor of aerospace and mechanical engineering. "This prize acknowledges that we have a credible supercomputing center at this institution."

Fasel heads the High Performance Computing Committee that directs supercomputing resource development at the UA and reports to Powell. The advisory committee formulates and recommends policy related to access, use and allocation of university supercomputing resources to optimize benefits for the university community.

Mathematician John von Neumann (1903-57) laid the foundations for scientific supercomputing in the 1940s. In recent years, supercomputing has been key in the human genome project, research on turbulence and other complex problems.

There are many different definitions for a supercomputer, but Fasel said one that he likes the best is that a supercomputer is "the most powerful computer available at a given time."

"Therefore, the supercomputer is a moving target, because time passes and technology changes," Fasel said. "What was a supercomputer 20 years ago is now a PC -- and what we call a supercomputer now is equivalent to the computing power the PC will have in another 20 years."

The UA ventured into supercomputing in the mid 1980s, when it acquired a "mini" supercomputer, an SCS40. It upgraded to more powerful Convex computers in 1989. They were powerful for their time, but "no competition for what we have now," said Kathleen Bowles, principal systems programmer for CCIT. "The Convex was still under gigaflops power. We wanted to break the gigaflops margin."

The university broke the gigaflops barrier when it installed the first SGI machine in 1997. Overnight, the university realized 30 times greater computing power. And a second upgrade, in December 1999, tripled computing power again.

A valuable part of Super includes graphics hardware for scientific visualization, said Mark Borgstrom, CCIT's manager of research computing. "One of the big advantages is that it includes an 'InfiniteReality2' graphics engine, which puts the graphics hardware for scientific visualization right on the same machine where the calculations are being performed."

Researchers need real-time animations and visualizations to make sense of the massive amounts of data that Super generates, Fasel explained. "Even if every resident of Tucson read data for a year, they would not be able to digest all the information generated by a supercomputer in a few hours. Graphics and other post-processing systems are necessary tools to extract the relevant mechanisms and advance science."

Fasel said he believes that a competitive supercomputer on campus is a must if the UA is to retain its Research I University status. "In the future, more and more research money from federal agencies will be earmarked for research that requires and uses high performance computing."

That's because federal research support is shifting from increasingly expensive, experimental research to less costly theoretical research and large-scale numerical simulations based on complex mathematical models and equations that can only be solved with supercomputers, he added.

"Since the overall research budget is not rising that much, if the UA isn't competitive in attracting those funds, then we will overall lose," Fasel said.

So what's the price tag for Super?

The cost of high performance UA computing has remained flat, or actually fallen, factoring for inflation, said Peter J. Perona, executive director of CCIT.

The university pays essentially the same for supercomputing power in the year 2000 as it did in the mid-1980s -- about $500,000 a year, Perona said.

The TOP500 list is online @ http://www.top500.org

(Editors: Michael J. Rule, editor, CCIT, contributed to and is a media contact on this story. Contact him at 520.621.5111, mrule@u.arizona.edu)

Share

Resources for the media