US Aims for Fastest Supercomputer Title
May 13, 2004 2:27 PM PT
The United States has set its sights on becoming the country with the fastest supercomputer. The Department of Energy's Oak Ridge National Laboratory (ORNL) announced it has been chosen to lead a partnership with the goal of building the most powerful supercomputer by 2007. The Department of Energy awarded the lab a US$25 million contract for the effort.
Development will be done by combining partnerships, researchers, new buildings and equipment. The blend of these different components is referred to as the National Leadership Computing Facility (NLCF). The five-year plan will pool computational resources for a sustained capacity of 50 teraflops (50 trillion calculations per second), with a peak capacity of more than 250 teraflops.
ORNL expects to reach the 45-teraflop mark by late this year or early 2005, according to Thomas Zacharia, ORNL associate laboratory director for Computing and Computational Sciences. He told the E-Commerce Times that the plan also includes hitting the 100-teraflop level in 2006.
Need for Speed
The push to craft a faster supercomputer is not simply rooted in a desire to find out how much computing power can be created, Zacharia said. Rather, speedier technology will aid in science and technology to such a degree that it will actually change how science is done.
"It's not just a big computer that will come out of this," he said. "It's the creation of a new national scientific fabric."
He noted that there are already fundamental changes in how sciences like nanotechnology and biotechnology are approached because of computing power. With even more supercomputing speed, Zacharia predicted, there will be breakthrough discoveries in climate forecasting, biology and fusion energy, among other fields.
"We will be able to investigate matter in a new way," he said. "It will allow us to understand the world we live in on a fundamental level."
The NLCF has signed up more than 20 partners for its supercomputing effort, including Argonne National Laboratory, which will develop the software that will run the supercomputer.
Argonne will install a 5-teraflop IBM BlueGene/L supercomputer that will be built at the lab's facilities. Rick Stevens, director of Argonne's Mathematics and Computer Science Division, said in an interview with the E-Commerce Times that the machine is scalable enough that the lab and IBM expect to be able to expand it to 40 teraflops.
"This project will set the direction of Argonne's high-performance computing research through the end of the decade," he said. "The BlueGene will allow us to explore a wide range of applications and promising technologies."
The NLCF also will increase the capacity of its current Cray X1 computer to 20 teraflops this year and will add a 20-teraflop Cray Red Storm in 2005.
America the Super
Currently, Japan holds the fastest-supercomputer title with its Earth Simulator, which has 40-teraflop capability. Zacharia noted that having faster supercomputing power in the United States not only will aid global scientific efforts, but also will be a boon for the country.
"As a nation, we have to innovate faster than the rest of the world," he said. "Clearly, this technology will have an impact for all nations, but it's in our best interest to have it here for economic competitiveness and faster innovation."
Aberdeen Group analyst Peter Kastner told the E-Commerce Times that such a supercomputer also would be handy for military applications, as it would eliminate the need for real-world testing.
Zacharia added that beyond scientific applications, such a powerful supercomputer also could be used for corporate strategies. For example, an automaker spends about US$2 billion to design a car and bring it to market. Using a supercomputer, design and production time could be reduced, lowering costs significantly. He maintains that this, in turn, would have a powerful effect on the nation's economy.
"It allows the country to expand its leadership position in the world and maintain it," he said.
Kastner added that the ORNL's goal is a realistic one that likely will succeed. "Even in the last five years, the low costs of microprocessors and the increasingly easy ability to scale computers has allowed the production of computers that were unforeseeable even 10 years ago," he said. "So, like a lot of things, as long as they've got the money, it's possible to develop the technology."