Supercomputers are, by definition, the fastest computers on the planet. The latest atop the Top 500 list, IBM's Blue Gene/L, can perform calculations at a blazing rate of more than 70 trillion a second.

But as processing speeds increase, so do the demands for larger, faster computer memories. And that's why organizers of SC2004, the largest supercomputing conference on the planet, have assembled in the David L. Lawrence Convention Center a data storage capacity that also is the largest on the planet.

Called StorCloud, it provides an unprecedented one petabyte of memory for people and organizations attending this week's conference. That's storage for more than 1,000 trillion bytes of computer information -- about 2 million times more capacity than the 50-gigabit hard drive that might be found on a home computer.

A petabyte is large even by the outsized standards of nuclear weapons laboratories, some of the largest users of supercomputers. Between them, Sandia National Laboratories' sites in Albuquerque, N.M., and Livermore, Calif., have maybe 200 terabytes -- 200 trillion bytes -- of storage capacity "and we are extremely blessed," said Ken Washington, Sandia's director of distributed information systems in Livermore.

A medium-sized university, by contrast, might have capacities of 10 terabytes or less.

Virginia To, a computer consultant from the Baltimore area who heads the StorCloud committee, said it took 32 tons of equipment worth about $80 million, donated by 22 vendors, to create StorCloud on the exhibition hall floor.

Consuming some 300 kilowatts of power, the storage units kick off enough heat to make StorCloud "the warmest place in the building," said Mike Knowles of Raytheon, who was in charge of the system design.

Creating StorCloud was a challenge in itself, but its larger purpose was to give those in attendance at SC2004 a chance to show how such a large memory capacity might be used, whether to perform protein sequencing, to calculate molecular energies using a quantum chemistry code or to simulate galaxy formation.

"We're trying to establish storage as one of the legs of high-performance computing," Washington said.

It is an idea heartily seconded by Robert Bishop, chairman of Silicon Graphics Inc., which joined with NASA this year to build Columbia, a 50-teraflop supercomputer that now ranks as the second fastest in the world and the fastest currently in operation.

Supercomputers are used primarily to perform simulations of a complex phenomenon, whether that be a nuclear explosion, an earthquake or the flow of blood through a coronary artery. But there's no way that those simulations can come close to nature if the computer can't hold all of the data it needs for a computation in its active memory, Bishop said.

"The jobs are just so big that we have to keep them in memory," he said. With computer speeds expected within two or three years to reach the petaflop level -- more than 1,000 trillion calculations per second -- "we've got to have petabyte memories."

Almost a dozen organizations at SC2004 brought applications with them to test on StorCloud. Investigators at four groups -- Sandia Labs, the San Diego Supercomputer Center, Pacific Northwest National Laboratory and Lawrence Livermore National Laboratory -- also are taking part in a challenge to determine who is most innovative in the use of the large memory.