National Supercomputing Center in Sweden uses SGI Altix new complex tasks
Swedish researchers have now the nation’s largest high-performance computer with shared memory architecture. It is a Linux-based supercomputer system SGI Altix 3700 Bx2, equipped with 64 Intel Itanium 2 processors and 512 GB of memory. At the University of Linkoping (LiU) in the National Supercomputing Center (NSC) set up computer platform from SGI (Silicon Graphics Inc., OTC: SGID), many physicists and other scientists from all over Sweden allow existing boundaries of computer-aided research to overcome and in areas such as materials science new highly complex issues to pursue.Already the first project, the NSC on the new high-performance computers will be realized, it could in the field of materials research revolutionary contribution. With the help of the SGI platform Swedish researchers to seek innovative methods to represent “organic electronics”, in which the potential of a cheap, easy alternative to existing manufactured silicon-based technology provides. The attention is the more accurate understanding of crystalline structures and the targeted exploring the possibilities of reliable electrical charge transfer in organic materials such outlet. In the course of the project are provided by the LiU researchers simulated Fehlordnungszustande and possible negative influences on the conductivity of the material analyzed. The computer simulation of such processes, they acquire new insights that will soon on the use of low cost processors in a variety of future applications could lead - of innovative electronic printed circuits to next-generation displays.
Shared memory processing - for controlled disorder
The physical simulation requires a coherent addressable main memory of at least 150 GB. Only then can the researchers correlated zufallsgenerierter many states with which they controlled the disorder in organic materials to describe try to calculate quickly enough. 150 GB shared memory, one of all processors uniformly usable main memory of this size, has none of the existing supercomputers and cluster systems at the NSC.
Although the Swedish researchers extrapolated a nationwide NSC cluster resource of a total of 1500 processors are available. And it has the largest NSC cluster would be a total memory of 400 GB. But the distributed memory architecture of the usual Linux clusters, the main memory accessible only distributed organized, it makes for such applications impossible to simulate the entire complex problem unzerstuckelt in memory to hold and efficiently as a whole.
Shared memory architecture - Study shows growing demand
Even in recent years, Swedish researchers to a previously installed SGI server system with 128 GB of shared memory use. The increasing needs of users and the fast-growing volumes of simulation models forced the NSC administrators, however, according to a new, more powerful computers look to keep the shared memory computing in a 64-bit Linux environment .. NPC Director Sven Stafstrom, professor of physics at the Computer Linkoping University: “Our primary need was a great shared memory platform. Stafstrom had in 2005 in a study about 50 user groups according to their needs and expectations of the High-Performance Computing (HPC) and specifically questioned the need for shared memory resources .. “The study showed that throughout the NSC network growing demand for such a super-computer type, because our cluster systems sufficient shared memory Missed opportunities. Around 20% of the surveyed researchers wanted access to a resource with a 200 GB contiguous usable main memory. Now we can offer them. ”
Optimal cost peak power for VASP, Gaussian and more
Four computer systems used in the selection came closer, has been approved by the NSC administrators over three weeks in intensive benchmark tests sent. Scientists examined the performance of the systems under its essential applications VASP and Gaussian and evaluated how well each of the scalable platforms in terms of online real application performance. The NSC decided on Altix as a super computer with the most balanced price-performance ratio. With the SGI system at the NSC under the operating environment ‘SuSE Linux Enterprise Server 9′ with ‘SGI Pro Pack ™ 4′ will work, researchers can quantenchemische problems in the application Gaussian up to 4x faster than solve before. “Not only the factor of price / performance was in the best Altix,” said Stafstrom. “SGIs technical staff has us very carefully done and help us competently helped the application potential of the software Gaussian better understand and make learning. This experience and the long-standing business relationship with SGI have convinced us that SGI us in the establishment and operation of the new system as a strong partner. ” In addition to VASP and Gaussian, the NSC researchers also DALTON quantum chemistry software on the new Altix system. In addition, Swedish scientists their own, often self-developed or further developed applications on the SGI machine can load as they did with the NSC clusters and servers are accustomed to.
Prospects for existing and new users
“Shared-memory systems can not only administer simple but often easier to use than cluster systems,” says Stafstrom. “We expect that some new users will be added when they see how easy it is to their software to Altix system and to put them there to work faster. We hope, Swedish scientists with the new supercomputer new perspectives for innovative solutions to tap. For new tasks, which they previously due to a lack of resources simply not tackle. ”
The purchase of the Altix machine was funding by the Swedish Research Council.
Flexibility - for today and tomorrow
Altix features for complex gleitkomma-intensive scientific applications such as Gaussian, because the platform for such tasks leading application throughput and leading scalability. This applies SGIs award-winning shared memory architecture NUMAflex (TM) (a cache-coharente NUMA system architecture already in 3rd generation) in combination with the leading floating-point performance of Intel Itanium 2 processors. The shared memory SGI NUMAflex architecture allows scientists, even the largest simulation models completely in the main memory to hold and edit, so they can data analysis faster and with greater interactivity, and ultimately to reach deeper-reaching knowledge and sustainable conclusions. In particular Altix systems offer a unique opportunity to all major types of system resources such as memory, general-purpose processors, accelerating special algorithm processors, I / O bandwidth, etc. independently scale - so that the system fit the current requirements for each configure and growing challenge hochrusten.
About Silicon Graphics GmbH:
SGI - Innovation for Results ™
SGI (NASDAQ: SGIC) offers a complete portfolio of high-performance server and storage solutions and industry-leading professional services and support services that help customers meet the challenges of complex, data-intensive workflows, faster innovation and discoveries as well as the Transformation of their information. Be it customers, the new active in the life sciences research, safer and more efficient vehicles and aircraft to develop the weather forecast and the global climate explore new technologies in mission-homeland security and defense, the switch from analogue to the broadcaster-band basierten on IT-based digital infrastructure, or customers, the company comprehensive records management. We are headquartered in Sunnyvale, California.
If you enjoyed this post, please consider to leave a comment or subscribe to the feed and get future articles delivered to your feed reader.
Comments
No comments yet.
Leave a comment