Quadrics and Bull have cooperated to develop high-end solutions for HPC clusters to bring unprecedented performance to those applications that require fast network communications. Outstanding scalability of the bandwidth On a three-rail QsNetII cluster of NovaScale servers a bandwidth of 2,541 GB/s has been measured using the PMB benchmark. Unlike other technologies, the latency of the network remains excellent even when all the nodes of the cluster are used, so that even with load-balancing Bull and Quadrics have been able to demonstrate latency under 5µs. Multi-rail configurations are required to achieve good scalability on applications with a high level of communications between the MPI processes. With a three-rail network, the scalability between 16 and 32 processes has been multiplied by 10 on the Amber benchmark. “We are very pleased to see that the unique multi-rail feature of QsNetII delivers a perfect scalability while keeping excellent latency. With Bull NovaScale servers clustered by Quadrics, customers enjoy a high-end solution delivering outstanding productivity for the most demanding applications” stated Cristoforo Romanelli, Managing Director of Quadrics. “Through our cooperation with Quadrics, Bull can provide solutions with unique performance levels resulting in quicker and more powerful simulations for our customers.” declared Jean-François Lavignon, Bull HPC director.