Mellanox, Lawrence Livermore National Laboratory Demonstrate Leading Performance and Scalability for HPC Applications

Continuous Collaboration Between Organizations Helps to Maximize HPC Clustering Efficiencies and Boost Application and User Workload Performance

Mellanox Technologies and Lawrence Livermore National Laboratory (LLNL) have announced world-leading scalability achieved with LLNL supercomputers and Mellanox InfiniBand interconnect solutions. The results are based on collaboration between the two organizations to enhance high-performance computing (HPC) software drivers and MPI libraries on top of Mellanox’s scalable interconnect solutions. The joint effort has delivered new levels of workload performance and maximized the return on investment for LLNL users.

“LLNL is tasked to take on some of the world’s most difficult and complex problems – all of which are very compute-intensive and place huge performance demands on the network,” said Matt Leininger, deputy for advanced technology projects at LLNL. “The collaborative work with Mellanox has helped us to recently advance our scaling capabilities in our large scale Hyperiontestbed cluster, delivering significant benefits on tested workloads. Deploying these enhancements on our production cluster will result in faster application runtime which increases research productivity.”

The Mellanox and LLNL teams modified Message Passing Interface (MPI) libraries to optimize Mellanox InfiniBand capabilities and enhance application performance and efficiencies. The tested workload performance at scale increased up to 700 percent and resulted in significant progress to meet LLNL’s petascale and exascale computing goals.

“Mellanox InfiniBand solutions deliver the highest scalability and performance for the supercomputers of today and the future,” said Michael Kagan, Chief Technology Officer at Mellanox Technologies. “The tight collaboration with LLNL enables both organizations to optimize application utilization over Mellanox InfiniBand solutions for the high-performance computing community, and to deliver enhanced performance and scaling for LLNL applications and workloads.”