MILPITAS, Calif. and TOKYO, Japan, June 19 — SGI Japan, Ltd. part of SGI, a global leader in high performance solutions for compute, data analytics and data management, announced that it shipped an order to The Institute for Solid State Physics (ISSP) of the University of Tokyo, for its next supercomputer system. The new system will be used by material science researchers across Japan and is scheduled to start operation on July 1 of this year. The theoretical calculation performance of the system is about 2.65 petaflops, approximately 14.5 times the performance of a conventional system, and it will be among the largest SGI systems in Japan.

Established in 1957 as a joint-use research institute and material science research organization, ISSP has provided scientists throughout Japan with supercomputing resources for computational research. As the development of calculation methods and demands of applied calculations in material science have become more diverse in recent years, significant calculation capacity is required. With the increased scale of calculation processes and rise in the number of users, ISSP chose SGI for the powerful system. The enhancement will include two sets of systems, including a cluster system with multiple nodes for parallel processing and a shared-memory system enabling high-speed processing in a single node. The new systems are expected to provide significant improvements in performance and scale.

“Like many research organizations today, ISSP has experienced an enormous influx in the data available to their scientists,” said Jorge Titinger, president and CEO, SGI. “Without SGI’s technology partnership enabling ISSP to process complex calculations in real time, researchers are greatly inhibited in their ability to make scientific breakthroughs. At SGI, we are proud that our systems are enabling researchers across Japan to utilize the massive amount of data available to them, enabling significant breakthroughs in material science. We look forward to continuing this relationship and furthering our high-performance computing leadership across the world with the technology needed to perform extremely data heavy and complex research.”

The clustered SGI ICE XA system will have a hybrid configuration of Intel Xeon processors and NVIDIA Tesla K40 GPU accelerators, a key part of the industry-leading NVIDIA Tesla Accelerated Computing Platform for accelerating data analytics, deep learning and scientific computing. The large-scale SGI ICE XA parallel computing system will feature 1,872 nodes, 44,928 cores, and 239.616 terabytes (TB) of memory with 576 Tesla K40 GPUs embedded in 288 of the 1,872 nodes. With a view to future exascale calculation performance, general-purpose computing on graphics processor units (GPGPU) are used for concurrent improvements in performance and restraint of power consumption. In addition, the rack containing the cluster system is stored in the latest, dedicated sealed cabinet, E-Cell, where a heat exchanger, developed by SGI, is employed to chill the entire rack with a water cooling system. Simultaneously, water-cooled heat sinks directly cool the CPUs and GPGPU, which further enhance the energy-saving benefits.

“System providers use Tesla accelerators to develop high-performance, energy-efficient computing solutions for a wide range of scientific and high-performance computing workloads,” said Roy Kim, group manager of Accelerated Computing at NVIDIA. “The GPU-accelerated SGI system will allow ISSP to provide its scientists with new levels of computational horsepower to accelerate their most challenging material science research.”

An additional 19 SGI UV systems with Intel Xeon processers embedded (760 cores and 19 TB memory in total) have been selected as the shared-memory system. The shared-memory system has 1 TB of shared memory per node and theoretical calculation performance of 31.6 teraflops. Together with the clustered SGI ICE XA, the new system is an ultra-high speed supercomputer capable of approximately 2.65 petaflops of performance.

By updating to the new SGI system, ISSP is promoting advanced research using calculations with a high degree of parallelism, development and spread of large-scale parallel software algorithms. The Institute’s aim is to extend exponential progress of computational physics and chemistry in material science and to construct and maintain the structure to support the development of computational science in Japan.

“Now, our Institute will have two sets of supercomputer systems, the SGI ICE XA and SGI UV, to enhance the joint-use service for material science researchers in Japan. Users all over Japan have used this service to contribute to the development of solid-state physics, which clarifies various characteristics of materials surrounding us based on physical laws,” said Osamu Sugino, associate professor, the Institute for Solid State Physics, University of Tokyo. “A higher calculation capability is required for research deployments with more complicated substances and high-affinity environments in order to conduct leading-edge studies using new algorithms. The new system responds to these requests by deploying Intel Xeon CPUs that users are familiar with and has a well-balanced combination of nodes with wide bandwidth and those with the calculation accelerator GPGPU. The GPGPU is installed for the first time in this Institute for improvement in electricity efficiency and popularization of heterogeneous computing in the study of solid-state physics. I am certain that this system will significantly contribute to computational material science in Japan.”

About SGI

SGI is a global leader in high performance solutions for compute, data analytics and data management that enable customers to accelerate time to discovery, innovation, and profitability. Visit sgi.com for more information.

—

Source: SGI