SCIENCE

Intel’s Discovery Running SWARM Also Achieves Second Highest Performance Per Core

E.T. International (ETI) has helped achieve four spots on the June 2011 Graph 500 List, including three in the top 10 and one with the second highest performance per core, as unveiled at the ISC’11 show last week in Hamburg, Germany.  Supercomputers using ETI’s SWARM (SWift Adaptive Runtime Machine) runtime environment came in at numbers six, eight, and nine for their performance on the Graph 500 benchmark.  

ETI’s SWARM delivered the best multi-node results for every scale problem to which it was submitted.  Specifically, when running SWARM, the Texas Advanced Computing Center’s (TACC) Lonestar 4 came in at number six.  At 512 nodes (Scale 34), Lonestar delivered 8.1 giga-edges per second (GE/s).  In the eighth spot, SWARM’s performance on Sandia National Lab’s Red Sky at 512 nodes (Scale 33) delivered 9.5 GE/s, almost double the speed of MPI on the same machine.  And at 256 nodes (also Scale 33), Intel’s Endeavor was number nine on the list when running SWARM, at 6.9 GE/s, an incredible 11-fold improvement over MPI.

Also using SWARM, Intel’s Discovery (Westmere E7-4870) placed 26^th on the list at a single node (Scale 27), and was one of three special mentions for its achievement of the second highest performance per core at 17.625 million traversed edges per second per core.

“These systems’ performance on the Graph 500 is a validation of the power of ETI’s technology to allow programmers to realize the capacity of advanced many-core computing systems,” said Rishi Khan, VP of R&D at ETI.  “SWARM’s impressive results underscore how critical it is to put the right runtime environment in place in order to exploit the parallelism of advanced machines.  The Graph 500 benchmark is one we take very seriously as an effective measure of a computer’s ability to efficiently solve the more advanced, data-driven problems that are so pervasive in today’s applications.”

ETI’s SWARM runtime environment was developed specifically to address the performance challenges presented by the intersection of massively parallel software applications and hybrid many-core computing systems.  The software enables easy and efficient programmability of applications for many-core computing systems, and is one of many components of ETI’s industry-leading system services, including benchmarking, hardware and software design, performance-critical programming, consulting, and more.