More significantly for NASA, Pleiades ranks 5th in the world and 3rd in the U.S. on the June 2015 High Performance Conjugate Gradient (HPCG) benchmark list, also announced at ISC. This "companion metric" to the LINPACK benchmark provides a more accurate performance measurement of the real-world scientific and engineering work done on general-purpose systems like Pleiades. This was the first time that Pleiades' performance numbers were reported on the HPCG list, which debuted in June 2014.
"Our goal continues to be to give scientists the computing capacity needed to do their research", stated William Thigpen, systems engineering branch chief at the NASA Advanced Supercomputing (NAS) facility at NASA's Ames Research Center in Moffett Field, California. "The impact of our continuous performance improvements to Pleiades is not about numbers on a list, but to support the ever-increasing modeling and simulation needs of missions across NASA - from aeronautics, to space exploration, to Earth and space sciences", William Thigpen stated.
For example, in May 2015, using data from NASA's Gravity Recovery and Interior Laboratory (GRAIL) mission to measure the moon's gravity field interior structure from crust to core, scientists running very large computations on Pleiades' Intel Xeon E5-2680v3 (Haswell) processors got results in half the time it took on the previous-generation processors. The improvement will save more than 2.5 million hours of computer time over the life of the project.
As another example, the increased number of Haswell nodes on Pleiades greatly reduced the time required to run roughly 10,000 cases for the booster separation aerodynamic database to support the first flight of Space Launch System (SLS) in 2018. The database is critical for SLS flight qualification, to determine the risk of the boosters recontacting the core after separation, which could cause loss of mission. The additional nodes cut in half the time required to populate the remaining data.
A series of upgrades over the past nine months increased the 211,360-core Pleaides' sustained performance to 4.09 petaflops (quadrillion floating-point operations per second) on the LINPACK benchmark. Sustained performance on the HPCG benchmark was measured at 131.90 teraflops (trillion floating-point operations per second). The upgrades included the addition of 4,176 12-core Haswell processors, each performing twice as many scientific calculations per second as the previous generation of Intel processors.