University of Cambridge Installs Mellanox FDR IB

Mellanox Technologies, Ltd that its FDR IB solution provides the University of Cambridge GPU-based supercomputer with scalable performance.

The system has a sustained performance of over 250 TF and ranked 166th on the November 2013 TOP500 list of supercomputers. The system was designed to increase energy efficient performance computing and ranked second in the November Green500 list that ranks supercomputers by energy efficiency.

The newly deployed supercomputer is partly funded by STFC to drive the SKA computing system development within the newly formed ‘SKA Open Architecture LAB.’ The SKA is a multinational collaboration which is building the world’s largest radio telescope which, at its center, is a requirement for the world’s largest streaming data processor, many times larger than the most powerful HPC system in operation today. The new system will take a central role driving system development for the SKA placing STFC and the University of Cambridge at the forefront of large scale big-data science.

“The network design was specifically architected to provide the highest I/O bandwidth for large scale big-data challenges and to have the highest message possible for large parallel application scaling,” said Paul J. Calleja, director of HPC at the University of Cambridge. “We chose Mellanox’s end-to-end FDR IB interconnects to connect the system, specifically by using their Connect-IB adapters in a dual rail network, as well as utilize the NVIDA GPUDirect RDMA communication acceleration to increase the systems parallel efficiency.”

“We are pleased to have Mellanox’s FDR IB solution as the interconnect of choice for the University of Cambridge’s supercomputer, the UK’s fastest academic-based supercomputer,” said Gilad Shainer, VP marketing, Mellanox. “Utilizing Mellanox’s Connect-IB adapters, the University of Cambridge is able to take advantage of the adapter’s leading message rate and bandwidth performance to enable fundamental advances in many areas of astrophysics and cosmology.”

Connect-IB is a scalable server and storage adapter solution for HPC, Web 2.0, cloud, big data, financial services, virtualized data centers and storage environments. Connect-IB adapters deliver throughput of 100Gb/s utilizing PCIe 3.0 x16, scaling with transport services, sub-microsecond latency and 137 million messages per second – 4X higher message rate over competing solutions.

Available, FDR 56Gb/s IB solution includes Connect-IB adapter cards, SwitchX-2 based switches (from 12-port to 648-port), fiber and copper cables, and ScalableHPC accelerator and management software.