Ryan Waite, product unit manager for Windows HPC Server, said it's important for Microsoft to have its software on the world's fastest computers.

"A lot of customers that are in the HPC space want to see that we can scale to very large systems before they'll actually deploy in smaller environments," he said from the SC08 supercomputing conference in Austin, Texas.

Seattle supercomputer maker Cray had the second-place system in the competition (as well as the seventh, eighth and ninth). The "Jaguar" at Oak Ridge National Laboratory reached 1.059 petaflops, just slower than the "Roadrunner," an IBM system at Los Alamos National Laboratory, which hit 1.105 petaflops.

The systems are measured using the Linpack benchmark application, a large linear algebra problem, much like the ones used in weather forecasting, engineering and other supercomputer tasks.

"These are the kinds of problems people encounter in mechanical engineering," Waite said. "Like if you're Boeing and you want to build an airplane out of something that's very complex to model, like carbon fiber, you really need to have a whole set of HPC systems behind you that are able to model all the stresses of the different components and compute the airflow as it goes over that system."

To reach 180.6 teraflops, the Dawning 5000A employed 1,920 servers each with four AMD quad-core processors. By comparison, a typical new desktop machine may hit 5 to 10 gigaflops, Waite said. One thousand gigaflops is one teraflops.

Microsoft launched Windows HPC Server 2008 in September. It's first product hit the market in June 2006.