By Scot Schultz
As supercomputing becomes more prevalent in university and academic research programs around the world, InfiniBand has emerged as a campus staple for accelerating these new clusters to achieve next generation compute power and advance applications such as artificial intelligence (AI), machine learning and more.
InfiniBand has long been a favorite of research institutions as they continue to develop supercomputers for a wide range of problem-sets. Universities have leveraged InfiniBand-based supercomputers to advance scientific programs that cover many applications – from molecular research at Kyushu University to weather and climate forecasting at University of Waterloo to research in scientific domains such as high energy physics, the human brain, renewable energy, and materials science at University of South Dakota.
Most recently, University of Michigan announced their next generation system called “Great Lakes”, and University of Toronto announced “Niagara” that have each been a world’s first in innovation with InfiniBand technology. The “Great Lakes” cluster is the world’s first system to deploy HDR 200Gb/s InfiniBand technology, while “Niagara” was also the first system in the world to deploy a world class system with dragonfly plus technology.
Demanding applications, such as AI and machine learning, require complex computations with extreme dependencies for fast data delivery. With a data rate of 200Gb/s, HDR InfiniBand provides the bandwidth to support in-networking computing memory for higher level communication framework collectives. In-network computing also dramatically improves distributed neural network training and overall machine learning application performance while eliminating the need for external parameter servers and reduces the amount of data that needs to move across the network which increases over-all system efficiency.
Universities and academic institutions choose InfiniBand as it meets both performance requirements, and is based on non-proprietary, open technologies. The InfiniBand Architecture Specification is an open standard developed in a vendor-neutral, community-centric manner. With a long history of addressing HPC and enterprise application requirements for I/O performance and scalability, the InfiniBand Trade Association (IBTA) and InfiniBand provides a reliable ecosystem for end users (such as universities) through open standards, roadmaps, compliance and interoperability testing programs, and educational resources.
Advances in supercomputing on campuses are leading to the use of HPC in many new and unconventional ways, such as high performance cloud computing by entities like Microsoft Azure. According to Brett Tanzer, Partner PM Manager, Azure Specialized Compute, customers are leveraging InfiniBand-enabled Azure to “push through new frontiers in high performance and accelerated computing, delivering the broadest range of Accelerated and high-performance computing (HPC) capabilities in the public cloud.”
InfiniBand continues to lead the TOP500 Supercomputer list and enhance industries and academic research further into the future through reliable speed and compute power.