SYSTEMS

PRIMERGY-based system to run on next-generation Intel Xeon processors: Fujitsu has announced that it has received an order from Japan's Institute of Physical and Chemical Research, known as Riken, for a new supercomputer system with a peak theoretical performance of 108 teraflops, approximately 9 times the performance of the existing system. The supercomputer will be a complex system comprised of three compute sub-systems. At its core will be a massively parallel cluster system consisting of 1,024 of the latest Fujitsu PRIMERGY compute nodes running on the next-generation Intel® Xeon® processors, code-named Nehalem. It is scheduled to begin operations in the middle of fiscal 2009. Background on New Supercomputer System As Japan's premier research institution in the natural sciences, Riken is pursuing research in a wide variety of areas including physics, engineering, chemistry, neuroscience, and life sciences. To support the progress of R&D in these fields, Riken has always utilized the optimum cutting-edge computing systems available. The current system employs a Linux operating system, grid and web technologies, as well as PC clusters, which have been traditionally used on a smaller scale in individual work divisions, to create a large-scale computing center considered both in and outside Japan as a model for next-generation computing centers. Riken has decided to upgrade its supercomputer system to support the enhancement of its research and development activities. The new system will be an advanced version of the existing system and will deliver the computational performance required for research applications in fields such as life sciences and physics, where computing needs continue to grow. In addition, the new system will feature greater user-friendliness and higher operating efficiency. Overview The supercomputer will be a complex system comprised of three separate compute sub-systems, each with its own specific purpose: massively parallel calculations, large-scale memory calculations, and multipurpose calculations. In addition, each compute sub-system will be connected to a common front-end system, and disk and tape systems. As the core system, the massively parallel Fujitsu PRIMERGY cluster system will be comprised of 1,024 of the latest Fujitsu PRIMERGY compute nodes (2,048 CPUs, 8,192 cores). The supercomputer's operating environment will integrate Parallelnavi, Fujitsu's HPC middleware, to create a seamless, one-system environment in which users are unaware of which compute sub-system they are accessing. A robust security environment will be designed to facilitate access to the system via the Internet. Features 1. Next-generation Intel Xeon processor to deliver high-speed performance The massively parallel Fujitsu PRIMERGY cluster system serving as the core computational system will deliver high-speed performance through the adoption of the next-generation Intel Xeon processor, code-named Nehalem. Built into the processor is the new Intel QuickPath Interconnect technology that provides high-speed connections between microprocessors and external memory, and between microprocessors and the I/O hub, thereby greatly enhancing overall system performance. Further, the power-saving Fujitsu PRIMERGY compute nodes will reduce the energy consumption of the cluster system to one-seventh the current system in terms of consumption per teraflop. 2. Fujitsu's Parallelnavi HPC middleware to enable high operability Fujitsu's Parallelnavi is a comprehensive HPC middleware providing all the software needed for a supercomputing environment, from application development tools required to build scientific computing programs, to job execution tools(program processing units), and operation management software. The main features include the following: * Program development and high-speed execution environments for scientific computing that optimize the hardware's capabilities, such as automated parallelization functions for multicore processors * A high-speed job entry function enabling up to 1 million jobs to be submitted. * A checkpoint restart function that reduces calculation reprocessing In addition, to enhance the operability of the complex system, Parallelnavi maintains file data uniformity and consistency for simultaneous high-volume access requests from the separate sub-systems. The theoretical performance is 12.8GB/S, eight times faster than the current system. Comment from Riken's Ryutaro Himeno, Group Director, Research and Development Group: "The existing Riken Super Combined Cluster was an R&D project in itself. I am very grateful for the proactive approach demonstrated by the staff from Fujitsu, which has resulted in high praise of the system." "For the new system, we took the design concept of the existing one, expanded and upgraded it to handle the predicted increase in demand from life sciences research as well as an increase in data from high-energy physics experiments. I am counting on the high reliability of Fujitsu's hardware and software and the continuing support of Fujitsu's staff. We look forward to contributing to further progress in science and technology and the development of human resources through this information infrastructure center." Research success using Riken's supercomputer system Riken has supported a wide range of research activities in natural sciences using its Fujitsu supercomputer systems. Yoichiro Nanbu, the co-recipient of the 2008 Nobel Prize for physics, used the system in April 2008 to continue research on his groundbreaking principle of symmetry breaking.