Sun Gets 400 Teraflops Supercomputing Deal with Galaxy Servers

Published: October 10, 2006

by Timothy Prickett Morgan

Sun Microsystems got its start as a Unix workstation company that sold a lot of workstations and then servers to academic and research institutions that needed a lot of computational capability. In the late 1990s and early 2000s, Sun's Sparc platform lost its competitive edge, and while Solaris was still popular for scientific computing, Linux had really come on strong. Then, Sun adopted Opterons and took Solaris open source, and said it was OK to use Linux and Windows. And now, the tide seems to be turning a little.

Sun announced last week that its Sun Fire "Galaxy" platforms will be at the heart of a 400 teraflops, $59 million supercomputer cluster that will reside at the Texas Advanced Computing Center (TAAC) at the University of Texas at Austin and that will eventually plug into the TeraGrid distributed computing facility that is being funded by the U.S. government's National Science Foundation. Arizona State University and Cornell University also have a hand in building and using the resulting machine.

TAAC is home to a Sparc-based supercomputing cluster called "Maverick," which is comprised of a single E25000 server with Solaris and Grid Engine grid software that is a visualization front end to the TeraGrid. The high memory bandwidth and large shared memory space of the E25K makes it a good--if somewhat expensive--machine on which to drive a lot of video cards for intensive graphics.

If the Maverick machine was a kind of frontal lobe for the TeraGrid, then the new 400 teraflops machine being installed at TACC is going to be a big piece of the cerebral cortex. (In fact, it will be the largest amount of computing in one spot on the distributed TeraGrid, which has many supercomputers hooked together so researchers can push workloads across a high-speed network to the most appropriate architectures to run their simulations.) The cluster will be comprised of servers using 13,000 of Advanced Micro Devices' future quad-core variants of the Rev F Opteron processors. These processors are expected sometime around the middle of next year. The machine will also have over 100 terabytes of main memory scattered across its machine, and 1.7 petabytes of storage.

In terms of raw computing capacity, this is the biggest HPC deal that Sun has taken down in its history, and the odds are that it will do many more such deals now that it has embraced the Opterons fully. Back in November 2005, Sun, NEC, and AMD partnered to build what will eventually be an 85 teraflops cluster for the Tokyo Institute of Technology, which at the time was the largest deal it had ever done. The initial cluster is a 50 teraflops machine that is comprised of 655 of the 16-core Sun Fire X4600 servers that were announced this summer, which works out to 5,240 Rev E dual-core processors. To boost number-crunching performance of the TiTech machine (for obvious reasons Tokyo Institute of Technology doesn't use a normal abbreviation), Sun added 360 floating-point accelerator cards from ClearSpeed.