Energy, DOD purchase big iron

The Energy and Defense departments are both installing new high-performance computer systems, the agencies have announced.

The Energy Department's National Energy Research Scientific Computing Center has taken delivery on a 722-processor cluster system from Linux Networx Inc. of Bluffdale, Utah. Running 2.2 GHz Opteron processors from Advanced Micro Devices of Sunnyvale Calif., this machine could offer a peak performance of 3.1 TFLOPS.

And while NERSC makes the final adjustments to this newly installed hardware, the Defense Department's High Performance Computing Modernization Program will take delivery on a new XD1 supercomputer from Cray Inc. of Seattle. Managed by the Naval Research Laboratory, this machine will run on 288 AMD Opteron dual-core processors and could perform at a peak rate of 2.5 TFLOPS.

Both machines will undertake computationally-intensive research and development work, though Energy's machine will do unclassified work while NRL's machine will run mostly classified jobs.

NERSC, located at Lawrence Berkeley National Laboratory in Calif., provides the high-performance computing for DOE's Office of Science, the basic research arm for DOE.

'Our mission is to provide leading-edge high-performance computing to the research community for the Office of Science,' said Bill Kramer, NERSC general manager

The NERSC machine'nicknamed Jacquard after mechanized loom inventor and punch card pioneer Joseph Marie Jacquard'will be available for Energy-funded researchers to run their jobs on remotely. It will support a wide range of projects, including climate modeling, material and life sciences, nanotechnology and astrophysics.

'This system will be used by basically every type of science application that there is,' Kramer added.

NRL's new XD1, to be located in Washington, will conduct mostly classified research work for Defense offices, according to Amar Shan, Cray product manager for the Cray XD1. The work will include crunching data for weather modeling, computational fluid dynamics, and chemistry and materials science.

What is unusual about Cray's machine is that it will use 144 Field-Programmable Gate Arrays, Shan said. Until recently, FPGAs were not widely used in high-performance computing work. Nonetheless, they can offer a distinct performance advantage, especially with low-level bit manipulation tasks such as cryptography, Shan said. FPGAs are relatively easy to reconfigure through software, and so can be customized to work on a particular application. They can approach speeds normally associated with a specialized chip dedicated to a particular application, Shan said. test tset tset test tset tset test

'For the right class of applications, the FPGA can give you one to two orders of magnitude speed [improvement] relative to an Opteron or Intel chip,' Shan said.

About the Author

Joab Jackson is the senior technology editor for Government Computer News.

Stay Connected

Sign up for our newsletter.

I agree to this site's Privacy Policy.