ALBUQUERQUE, N.M. — Achievement of the one-trillion math-operations per second computing milestone ushers in a new era in which high-fidelity 3-D simulation will enable scientists to reach the eventual goal of preserving a safe, secure and reliable nuclear deterrent without underground testing.
Intel Corp. computer engineers demonstrated the one-teraflops breakthrough, considered to be the Holy Grail of high-performance computing, to officials from the Department of Energy and Sandia National Laboratories on Dec. 11 in Beaverton, Ore. The Intel massively parallel computer was developed under direction of the Department of Energy for the Accelerated Strategic Computing Initiative (ASCI), a 10-year program designed to move nuclear weapons design and maintenance from a test-based to simulation-based approach.
The Tflops, now the fastest supercomputer in the world, and which DOE, Sandia and Intel are touting as the world’s first “UltraComputer,” is being moved in sections from Oregon to Sandia over the next few months. The full system will consist of 76 large computer cabinets, with 9,072 Pentium Pro processors and nearly six billion bytes of memory. It will cover about 1,600 square feet, enough to fill a moderate-sized home. The one-teraflops demonstration was achieved using 7,264 Pentium Pro processors in 57 cabinets. The run took an hour and 20 minutes, and during that time, the machine performed 6.4 quadrillion floating point calculations.
“This awesome achievement marks the beginning of a new era in computing,” said Art Hale, teraflops technical manager at Sandia. “It opens the door to dramatically more realistic three-dimensional simulations, which will profoundly change the way we use simulation in science and engineering.”
Sandia computational scientists already are testing software applications that are key to ASCI’s success on a five-cabinet system assembled recently in a highly secure area of the national security laboratory. Sandia computational scientists are world leaders in the development of software applications for massively parallel supercomputers. Sandia and the other two DOE weapons labs, Los Alamos and Lawrence Livermore national laboratories, are refining existing applications and developing new applications tailored to terascale computing capabilities for ASCI. The computer, which is ten times, and in some applications 100 times, more powerful than the fastest machine used today, will have a peak performance capability of about 1.8 teraflops, or 1.8 trillion floating point operations per second. It would take someone operating a hand-held calculator about 30,000 years to calculate a problem the teraflops computer could compute in one second.
“The outstanding applications software development skills of Sandia and our DOE partners complement Intel Corp.’s superb computer hardware capabilities to create a dynamic combination that promises to revolutionize computational science in many disciplines,” said Bill Camp, director of Computational Sciences, Computer Sciences and Mathematics at Sandia.
The DOE, Sandia and Intel jointly announced the $55 million teraflops development contract in September 1995. The computer represents the initial goal of ASCI, which could culminate in computers with hundreds of teraflops capabilities by 2005. Computers that powerful are needed to simulate the complex 3-D physics involved in nuclear-weapon performance, and to accurately predict the degradation of nuclear weapons components as they age in the stockpile. Powerful multi-teraflops computers also will permit analysts to quickly run full-system 3-D simulations of complex accident environments, such as an airplane crash followed by a fuel fire, to predict safety against accidental nuclear explosions.
“Teraflops computing and ASCI provide an extraordinary opportunity for the three weapons laboratories in DOE to work together on behalf of the science-based stockpile stewardship program,” said Sandia Director C. Paul Robinson. “It is a very important step in shifting from a test-centered program to a computational-centered program.”
ASCI will integrate the applications codes and ever-increasingly powerful supercomputers. Even larger multi-teraflops machines will be situated at Los Alamos and Lawrence Livermore national laboratories over the next few years into a robust problem-solving environment by developing tri-lab secure high-speed networks, high-speed high-capacity storage facilities, parallel programming tools, and data visualization tools.
Sandia computer scientists have run applications on high-performance Intel computers over the past few years in a race with Japan to set computing speed records. Sandia/Intel and Japan continued to leapfrog each other’s efforts until a team of scientists from Sandia and Intel achieved a computing speed of 281 gigaflops in December 1995 by linking two Intel Paragon computers. That record was promptly surpassed earlier this year by a special-purpose Japanese system’s 368 gigaflops mark, a record which stood until this month’s achievement of 1.06 teraflops.
Sandia has played a fundamental role in the brief history of high-performance computing, particularly in the development of parallel supercomputing and its applications. Sandia first captured national attention for its work in massively parallel processing in March 1988 when it won two supercomputing prizes: the Karp Challenge for demonstrating unprecedented speedups using processors working together compared to processors running separately; and the Gordon Bell Prize for achieving a thousandfold speedup on three engineering problems analyzed with 1,024 processors working in parallel. Until the breakthrough by Sandia, most computer scientists believed using even thousands of processors could speed up problem-solving by no more than 50 to 100 times the rate of a single processor. Over the years, Sandia has demonstrated a particular expertise for developing mathematical methods, algorithms, and software that are required for large-scale parallel processing.
Among Sandia’s most notable software applications developments, all of which will run on the teraflops and perform key ASCI work, are CTH code used for computer modeling of high-speed impacts and the effects of nuclear explosions; a software package called CHACO that provides the means to make massively parallel computers easier to use by facilitating the mapping process that distributes computations across multiple processors; the PRONTO code, which can calculate structural mechanics and applied force, such as the crushing of a car or airplane in a crash or the deformation of a weapon upon impact; and ALEGRA, a series of codes that combines shock codes, such as CTH, with structural analysis codes.
Sandia’s massively parallel quantum chemistry programs have been applied to several real-world problems in medicinal chemistry, including studies of anticancer drugs and environmental carcinogens. And, Sandia scientists have used massively parallel machines to create 3-D models simulating underground formations for the gas and oil industry.