Fujitsu, JAEA unveil new supercomputer

Fujitsu said that it has completed joint development of a new supercomputer system with the Japan Atomic Energy Agency (JAEA), which can be used in various areas of atomic energy research including nuclear fusion simulations.

According to Fujitsu, the new supercomputer is a hybrid system that consists of three computational server systems such as large-scale parallel computation unit, application development unit for the next generation supercomputer and SMP Server (shared memory server).

Fujitsu said that the large-scale parallel computation unit, which forms the core of the new system, is able to deliver a high-performance parallel computing environment through the use of the company's new blade server Primergy BX900, in a configuration of 2,134 nodes (4,268 CPUs, 17,072 cores) connected using the new InfiniBand QDR high-speed interconnect technology.

In addition, the NxG code development unit uses a 300-node FX1 server cluster and the shared memory server uses a single-node SPARC Enterprise M9000 Unix server. There is also a 36-unit Eternus DX80 disk array storage system.

All three server systems run Fujitsu's high-performance computing middleware Parallelnavi, which delivers a common development and execution environment and unified operations management, the company said.

The company said that by mounting two InfiniBand QDR cards in each blade server one-way data transfers as fast as 8GBps can be achieved. The Primergy BX900 blade server is suitable for a range of applications, including enterprise computing systems, server consolidation and the building of cloud environments.

Toshio Hirayama, director of the centre for computational science and e-Systems of the JAEA, said: "Supercomputers are indispensible for the kind of scientific computations required in nuclear energy research and development. For that reason, I'm confident that the new supercomputer system will make it possible to bring calculations that had been impractically large within reach. By leveraging this new system, we intend to develop codes for the next-generation supercomputer that will be deployed in 2012."