INDUSTRY

Using a powerful new Cray XT3 supercomputer, the Department of Energy's (DOE) Oak Ridge National Laboratory (ORNL) has run the world's fastest, most-detailed simulation of waves used to control plasma -- gaseous matter superheated enough to generate massive amounts of energy. The simulation is related to the multibillion-dollar ITER project, which aims to tame plasma so it can later become a virtually inexhaustible supply of electricity. ORNL's new Cray supercomputer also posted leading results on the HPC Challenge standard tests of supercomputer performance, Cray Inc. reported today. En Route To Reproducing the Sun's Energy "Fusion researchers are acutely aware of Earth's dwindling supplies of petroleum and other non-renewable energy sources. They have been trying for decades to tame fusion, the process that powers the sun. Energy from fusion is released when hydrogen ions are heated to tens of millions of degrees," said Don Batchelor, head of the Plasma Theory Group in ORNL's Fusion Energy Division. "A major advantage of plasma energy, aside from its near-limitlessness, is that it does not contribute to global climate change." But harnessing plasma energy depends on scientists' ability to understand and control the behavior of the plasma. ORNL's record-setting simulation of plasma control waves promises to move this understanding an important step forward. ORNL performed the simulation under the DOE's SciDAC (Scientific Discovery Through Advanced Computing) program. "Using 3,072 processors, or about 60 percent of the new Cray XT3 supercomputer at ORNL's National Center for Computational Science, we were able to run the largest, most-detailed simulation ever done of plasma control waves in a tokamak, the donut-shaped reactor that will eventually form the core of the multinational ITER reactor," said Fred Jaeger, the ORNL researcher who ran the simulation. "We were able to complete the simulations three to seven times faster than when we run the same AORSA (All-Orders Spectral Algorithm) application on DOE's 'Seaborg' high-performance computing system." The U.S. has joined forces with a number of nations in Europe and Asia to develop ITER, the multibillion-dollar International Thermonuclear Experimental Reactor, by 2016. ITER is the experimental step between today's studies of plasma physics and tomorrow's electricity-producing fusion power plants. ITER's donut-shaped reactor will use magnetic fields to contain a roiling maelstrom of plasma that will serve as the 'fuel' for a massively energetic fusion reaction. ITER will operate at 100 million degrees Centigrade and generate 500 megawatts of fusion power. Record Results on HPC Challenge Benchmark ORNL's new Cray XT3 supercomputer has also posted leading results on the industry-standard HPC Challenge benchmark tests sponsored by the National Science Foundation, U.S. Department of Energy and DARPA HPCS (High Productivity Computing Systems) program. The HPC Challenge benchmark suite "is a better indicator [than any single test] of how an HPC system will perform across a spectrum of real-world applications," according to Jack Dongarra, who assembled the test suite with colleagues from the University of Tennessee and elsewhere. The HPC Challenge test suite is designed to provide an overview of the major strengths and weaknesses of various HPC systems. To ensure that system purchasers and others see the whole picture, the HPC Challenge website requires results to be posted for all of the tests. Computer-makers typically are reluctant to post a mix of good and poor results. The most recent HPC Challenge results from ORNL, run on 5,200 processors of the lab's Cray XT3 supercomputer, set a new record for the largest system to post results for the tests, surpassing a 3,744-processor version of the same Cray system at ORNL. This is important because the larger the system, the harder it is to achieve outstanding results on every test. The Cray XT3 system posted superior performance on four of the five global tests in the HPC Challenge suite: high-performance Linpack (G-HPL), global bandwidth (G-STREAMS), G-PTRANS and G-FFTE. On these important tests, which measure performance across the entire system, the Cray XT3 supercomputer's absolute performance was 4 to 13.6 times faster than the next-fastest listed U.S.-made system. "These outstanding large-scale results show that the Cray XT3 supercomputer is keeping its promise to deliver highly scalable application performance," said Peter Ungaro, president and chief executive officer of Cray. "The Cray XT3 system, with its balanced massively parallel processing architecture, was designed specifically to meet our customers' needs for superior performance when running large, complex applications across hundreds or thousands of processors. We are excited to be working with ORNL to break new ground in science and computing."