AEROSPACE
Cornell University Selects Topspin for High-Performance Computing
Ithaca, New York and Mountain View, Calif. - Topspin Communications, the leader in switched computing, and Cornell University today announced that Cornell has selected Topspin to provide InfiniBand hardware and software in support of a high-performance computing cluster being built by the Cornell Theory Center to showcase the performance potential of the Intel 64-Bit Architecture (IA64) and Windows 2003 server over InfiniBand. The cluster will be deployed in stages, beginning next month with an initial 16 node mini-cluster and scaling to much higher numbers of nodes throughout the year. Topspin will provide the InfiniBand infrastructure to interconnect the IA64 servers in the cluster including switches, host channel adapters, and the necessary system, subnet management, and host driver software. “We selected Topspin because of the completeness of their solution set,” said David Lifka, Chief Technical Officer, Cornell Theory Center. “Topspin was the only InfiniBand systems provider that could deliver both the hardware and software we require to support the message passing interface (MPI) over Windows 2003 and IA64 to meet our aggressive timelines. We are pleased to be working with a proven market leader.” “We are very pleased to be selected by Cornell for this exciting endeavor,” said Stu Aaron, VP Marketing and Business Development at Topspin. “InfiniBand is proving to be a great enabler for high-performance computing clusters that deliver both the performance previously reserved for mainframes and the superior economics of industry-standard components. We look forward to working closely with Cornell on this initiative.” InfiniBand is the only industry-standard, high-performance interconnect technology available today that provides an unmatched combination of 10Gbps high throughput, extremely low latency (less than 7 microseconds), and Remote Direct Memory Access (RDMA). This combination is critical for building high-performance computing clusters capable of delivering the performance of traditional supercomputers at a fraction of the cost.