The U.S. Department of Energy's Argonne National Laboratory will play a major role in the development of a new national computational science facility aimed at deploying a supercomputer capable of sustained performance of 100 trillion floating-point operations per second (teraflops) by 2007.
Argonne partnered with Oak Ridge National Laboratory to develop a proposal for the National Leadership Computing Facility (NLCF), which will engage teams of researchers from national laboratories, research institutions, computing centers, universities and vendors. The U.S. Department of Energy (DOE) will use the NLCF for mission-related research; the NLCF will be open to researchers from around the world for competitive, peer-reviewed research.
The Oak Ridge-Argonne proposal was selected from four proposals received from DOE's science laboratories. Energy Secretary Spencer Abraham announced May 12 that the department would provide $25 million to fund the project's first year. The total cost is estimated to be $150 million to $200 million.
“This world-class computational facility will ensure America 's leadership in high-performance computing and scientific discovery,” Abraham said. “It will serve to revitalize the U.S. effort in high-end computing.”
Oak Ridge National Laboratory will be responsible for working with vendors and users to determine the best system architecture for the expected set of computation problems. It will work closely with Cray and IBM as well as with Argonne, other DOE national laboratories and universities to make the new DOE computing capability a success.
Argonne will help develop software for two computers to be deployed at Oak Ridge National Laboratory: Red Storm, a massively parallel machine, and a Cray X-1 vector architecture parallel computer. The Cray will be upgraded to a more powerful version, the X-2, which is expected to achieve the 100-teraflops mark.
In addition, Argonne will deploy and evaluate, in partnership with IBM, a five-teraflops BlueGene/L supercomputer at the laboratory's Illinois site. Researchers at Argonne 's Mathematics and Computer Science Division will evaluate the system and the software needed to support scientific applications.
IBM's BlueGene system was selected to expand the range of computing architectures available to scientists and to explore promising technologies for next-generation leadership-class machines, those with performance in the petaflops range (1 petaflops = 1000 teraflops). Argonne leads a consortium of research institutions working on open-source software for this architecture.
“Research applications suitable for BlueGene supercomputers range from simulations of brain seizures to the processes in exploding stars,” said Argonne Director Hermann Grunder. “High-end computation, theory and experimentation are the three pillars supporting today's science and technology research. The laboratory is proud to be an integral part of this next major step in the evolution of scientific computation.”
“The BlueGene computer system, with its Power-based architecture, offers a great computational platform from which to search for new scientific breakthroughs,” said Pat Toole, general manager of IBM Engineering and Technology Services.
BlueGene system architecture is designed to be highly “scalable” — additional central processing units, or CPUs, can be added to the system with minimal changes to the operating system, applications and software environment. Argonne researchers will explore algorithms and techniques for achieving a larger fraction of peak performance, and will use both standard performance benchmarks and actual application codes to help predict how the system will perform as additional computing power is added to the system. Future versions of the BlueGene/L system may have more than 120,000 CPUs.