Efficient Supercomputing

The NAS Division set a new standard with the installation of Pleiades, one of the most energy-efficient supercomputers in the world. Pleiades ranks high on the Green500 supercomputer list at #22, based on computational efficiency—and when combining energy efficiency and computational power, Pleiades comes in at #2 (233 megaflops per watt) among the world's general-purpose supercomputers.

The system's state-of-the-art power supplies are at least 89% efficient, with only one per rack (plus a back-up), rather than dozens (one per node), which substantially reduces power losses. As a result, Pleiades uses a total of 2.09 megawatts of peak power—compared to other top supercomputers requiring up to 7 megawatts peak. The new technologies used in Pleiades make it about 4 times more efficient than the Columbia supercomputer, which went into production at NAS in 2004.

Green Planning

From the beginning of the procurement process, plans to acquire NASA's newest supercomputer emphasized savings in energy consumption and cost. The selection team tripled the weight applied to the "power usage" factor for the supercomputer. They chose an innovative system configuration—saving vast amounts of energy consumed and physical space—from a vendor with a stated commitment to following environmentally sound operations and manufacturing processes. Additionally, NAS limited memory to the amount necessary to serve the majority of users. Although a few applications see a slight reduction in performance per processor core, Pleiades' total power usage was reduced by 10-15%.

Green Practices

NAS continually seek ways to maximize the productivity of both the high-end computing (HEC) systems and scientific users, to get the most value out of the energy consumed. Among our green practices:

• System engineering methods to ensure processors not in active use automatically idle at low power
• Hardware upgrades and tracking tools to increase system utilization and reliability, reducing wasted computer cycles
• Application optimization tools and methods for increased efficiency—producing more computational results with the same, or fewer, resources
• Exploring GPU-based acceleration for dramatically increasing efficient computing for a few heavily used applications
• Transitioning to a more efficient approach of cooling supercomputers locally (using water-cooled doors), rather than cooling the ambient air

Whole Systems Perspective