Sun Readies a Supercomputer in Texas

The new supercomputer at the Texas Advanced Computing Center will handle some of the biggest problems in physics, astronomy and medicine.

If everything is bigger in Texas, then the new Sun supercomputer at the Texas Advanced Computing Center should fit right in.

Sun Microsystems and the University of Texas, where the TACC facility is located, will unveil the new Constellation system at a Feb. 22 ceremony in Austin. The new system, dubbed "Ranger," was made possible through a $59 million grant from the National Science Foundation.

At its peak, the Ranger machine offers 504 teraflops, or 504 trillion calculations per second. Of the world's Top 500 supercomputers, IBM's Blue Gene/L system, which is installed at the Department of Energy's Lawrence Livermore National Laboratory, is currently listed as the world's top-performing supercomputer with a maximum performance of 478.2 teraflops.

When fully functional, the Ranger system will work on some of the most important problems in the worlds of physics, astronomy, geology and biology.

Doug Toussaint, a professor of physics at the University of Arizona, is part of a team studying quantum chromodynamics, the theory of strong interactions to describe the relationship between quarks-the fundamental particles that create matter-and gluons, which are the elementary particles that allow quarks to interact with one another.

While Toussaint and his colleagues have only run limited experiments with the Ranger system to this point, they said they hope to eventually use the supercomputer to run simulations that will answer some of the fundamental questions of particle physics, such as by determining the masses of different types of quarks.

To develop better data, Toussaint and other researchers need access to high-performance machines with much more computing power than those they currently use.

"You are always going to have some error just coming from your limited statistics, and generally the size of these errors goes as one over the square root of the number of the samples you can afford to generate," Toussaint said. "So, if within all of these statistical errors, you get four times the amount of computing power that you had before, [the number of errors in the data decreases] by a factor of two. ... This is very exciting."

The next step for Toussaint and his team is testing different compilers to see how the Ranger scales with different combinations of processors, and then within a couple of weeks bring real production data applications to work with the supercomputer.

For Sun, the unveiling of Ranger marks a forceful return to the field of supercomputers and HPC (high-performance computing). For years, this particular part of the market has been dominated by IBM and its line of supercomputers, including its Blue Gene/L systems. Now, both Sun and IBM are looking to develop machines that will pass the petaflop mark, or 1 quadrillion calculations per second. The Constellation systems are Sun's first step in that direction.

The completed Constellation includes 82 Sun Fire blade server racks that contain nearly 4,000 modules, a pair of Sun Magnum ultradense switches, an InfiniBand host interface with 288 ports and next-generation Mellanox Technologies HCA (high-contrast addressing). The Ranger supports up to 123TB of memory and offers up to 1.7 petabytes of storage.

The unveiling is also a chance for the IT watchers to get a close look at Advanced Micro Devices' quad-core Opteron processors, which have been beset by problems since the company introduced them to the market in 2007. The Feb. 22 event is a chance to preview the processors in a large-scale deployment before other server vendors begin offering the quad-core Opterons in mainstream server systems.

The TACC system will support a total of 15,744 Opteron processors.

The AMD processors that Sun and TACC are using for Ranger still contain a bug the company described in December that involves the chip's translation-lookaside buffer, which caused problems for data being transferred from the Level 2 to the Level 3 cache. AMD announced that a BIOS patch would fix any problems, and Sun and the university's engineers applied a patch to Ranger to ensure that the system would work.