NASA Ames Installs World’s First 512-Processor Altix Supercomputer

NASA’s high-performance computing capabilities have taken a giant step forward with the installation of the world’s first 512-processor SGI® Altix™ single-system image (SSI) supercomputer at NASA Ames Research Center, Moffett Field, Calif.

NASA Ames, a renowned leader in the development of large single-system image machines, has been collaborating with SGI over the past seven years in the development of the world’s first 256-, 512- and 1,024- processor global shared-memory systems. Recently, engineers at NASA and SGI worked together to expand the capabilities of the SGI Altix line of scalable systems. The latest result is NASA’s new 512-processor SSI Altix system based on the Linux® operating system, the first of its kind in the world.

“With the addition of the new SGI Altix system, NASA’s high-end computing testbed activities in support of the agency’s science and engineering missions will be greatly enhanced,” said Dr.Walt Brooks, chief of the NASA Advanced Supercomputing (NAS) Division at NASA Ames Research Center, located in California’s Silicon Valley. “Thanks to its outstanding performance capabilities, this new testbed is already helping NASA achieve breakthrough results to meet major challenges in atmospheric and ocean modeling and aerospace vehicles,” Brooks added.

The SGI Altix 512-processor system is part of an ongoing NASA effort to push the limits of high-performance computing. “Creating and using this system took just a matter of weeks-from September to October,” said Jim Taft, lead for the advanced computing technologies effort at NASA Ames. “The current system is already in partial production, running mission-critical computations in aeronautics and Earth sciences around the clock. With the current workload, stability of the system has been excellent,” Taft said.

According to Bob Ciotti, an Ames research scientist and the lead for the center’s Terascale Applications Group, the new supercomputer achieved a Linpack Rmax rate of 2.45 teraflops and a STREAMS Triad rate of 1.007 terabytes per second-the fastest performance measurement in the world by both ratings for a shared-memory system, and the first to break the one terabyte limit on the memory bandwidth benchmark.

“Shared-memory systems have the communication characteristics necessary to scale applications to hundreds of processors,” Ciotti explained. “With this new Altix, the worst-case communication latency is less than a microsecond-and that’s important for sustained performance when running on all 512 processors.”

The new supercomputer is being used for a joint effort by NASA Headquarters, NASA’s Jet Propulsion Laboratory, Pasadena, Calif., NASA Ames Research Center and NASA Goddard Space Flight Center, Greenbelt, Md., to deliver high-resolution ocean analysis in the framework of the ECCO (Estimating the Circulation and Climate of the Ocean) Consortium, which involves the Jet Propulsion Laboratory, the Massachusetts Institute of Technology, Cambridge, Mass., and the Scripps Institute of Oceanography, La Jolla, Calif.

“The turnaround time users typically see on large ocean simulations can take months,” Ciotti said. “By dedicating half of the machine to the ECCO project and scaling the code to run efficiently on all the processors, we now expect turnaround to be about two to three days.” Scientists of the ECCO Consortium believe that running in this very high-performance environment will help them better understand ocean circulation and its impact on global climate patterns.

The new SGI Altix SSI system is both faster and more efficient than its predecessor, which was the world’s first 1,024-processor SGI Origin-based supercomputer, acquired by NASA Ames in November 2001. The Altix system also provides dramatically better price/performance, utilizing both the Intel® Itanium 2® processors and a 64-bit Linux operating environment.

NASA Ames’ quest for building increasingly larger SSI systems is driven by the goal of providing the simplest and most efficient system for high-performance computing to its scientific users. This simplicity in design has translated into rapid advancement through each stage of the project.

“As we progressed through each stage of development, from 64, to 128, to 512 processors, the system has performed almost flawlessly, with performance numbers routinely three to four times the previous best results at NAS with a similar number of processors,” said Taft.

The new Altix system is driven by Intel® Itanium 2® processors and has a total memory of about one terabyte. Introduced in January 2003, the Altix 3000 systems incorporate the high-performance SGI NUMAflex global shared memory architecture. The NUMAflex™ design enables the CPU, memory and its operating systems, graphics and storage to be packaged into modular components, or “bricks.” The new supercomputer is the first in a series of high end computing testbeds driven by a recently forged partnership between NASA’s Office of Aerospace Technology and Office of Earth Science.