Cray, a name long synonymous with the term supercomputer, has recently announced the availability of a new Linux cluster capable of putting out 691 gigaflops at peak performance in some configurations. The XD1 system runs on Opteron processors and uses a direct connect processor architecture to help it avoid bottlenecks and memory contention.

Part of the appeal of the new system is expected to be its cost. The US Forest Service, for example, was able to purchase one of the new systems:

“Tracking a smoke plume as it moves downwind from a fire requires all the computational complexity of a weather model run over a nationwide domain,” said Bryce Nordgren, a scientist with the Forest Service’s Fire Science Lab. “Tracking the evolving chemical composition of said plume produces a task so computationally intense that we assumed we would not be able to afford any computer capable of performing it,” he said in a statement.

Personally, I would like to see more breakdowns of where the cost savings for these systems are coming from. Is it commodity hardware? Is it the avoidance of hefty per-processor fees for proprietary software? Hopefully some future article will address this.