GOVERNMENT

Swedish researchers today announced they have installed the country's largest shared-memory supercomputer, a new 64-processor SGI Altix system equipped with half a Terabyte of memory. Deployed this month at the National Supercomputing Center (NSC) at Linkoping University (LiU), the new supercomputer from Silicon Graphics will allow physicists and other researchers from throughout Sweden to break through computational barriers created by complex computations. The first project slated for the NSC's new SGI Altix 3700 Bx2 may help researchers revolutionize the manufacture of "organic electronics" that could serve as a low-cost, easily manufactured alternative to silicon. The more LiU researchers can learn about the crystal structure of such systems, the better they can assess the ability of organic materials to reliably carry electrical charges. The project aims to simulate disorders that can hinder an organic material's conductive properties. Accurately simulating these properties will give scientists insight into the potential for use in the kind of low-cost processors found in a growing array of applications -- from printed electronics to next-generation displays. To quickly compute a large set of correlated random numbers that describe the disorder in organic materials, NSC's computational physics simulation requires 150GB of globally addressable memory, an amount no existing NSC supercomputer or cluster could provide. Throughout Sweden, researchers tap into a broad network of NSC clusters that together offer a total of 1,500 processors. Although the largest NSC cluster boasts 400GB of memory, the distributed nature of traditional Linux clusters makes it impossible for applications to efficiently hold the entire problem in memory. In recent years, Swedish researchers used a previously installed SGI server with 128GB of shared memory. But escalating needs and rapidly growing data sets prompted NSC administrators to seek out a new, more powerful shared-memory supercomputer that uses a 64-bit Linux operating environment. "Our primary need was for a large, shared-memory system," said Sven Stafstrom, NSC director and professor of computational physics at the University of Linkoping. Last year, Stafstrom surveyed nearly 50 user groups throughout Sweden on their high-performance computing (HPC) needs, and particularly on their need for a shared-memory resource. "The survey showed that throughout the NSC network, there is growing demand for such a supercomputer, because our cluster systems lack shared-memory capabilities. About 20 percent of the researchers we surveyed said they wanted to access 200GB of shared-memory right away. Now, for the first time, we can provide that." NSC administrators put four competing systems through three weeks of benchmark tests. Scientists tested system performance on VASP and Gaussian, and gauged the scalability of each platform in terms of real-world application performance. Working with funding from the Swedish Research Council via the Swedish National Infrastructure for Computing, the NSC selected the system that best balanced price and performance: an SGI Altix supercomputer powered by 64 Intel Itanium 2 processors and featuring 512GB of memory. The Altix system also uses Novell SUSE Linux Enterprise Server 9 with SGI ProPack 4. With the new Altix system, NSC scientists saw Gaussian problems run up to four times faster than ever before. "Not only was price/performance of the Altix superior, but we also received very attentive assistance from SGI's technical staff, whose expertise actually helped us learn more about the nature of Gaussian as an application," said Stafstrom. "This experience, along with our long-standing relationship with SGI, proved that SGI would continue to serve as a strong partner as we moved forward with the new system." In addition to VASP and Gaussian, NSC researchers will install DALTON quantum chemistry software on the new SGI Altix system. Swedish researchers will also be able to load their own applications onto the system, as they currently do on NSC clusters and servers. "Shared-memory systems can be easier to administer and use than clusters, so we expect that some new users may find it easier to get up and running faster on the Altix system," said Stafstrom. "We hope that with this new supercomputer, Swedish researchers will see the possibility to invent new problems that they have never even considered, simply because they did not have the resource to handle them." The SGI Altix platform excels at complex, floating-point intensive scientific applications such as Gaussian because of the Altix system's leading application throughput and scalability. This is enabled by the superior floating-point performance of Intel Itanium 2 processors coupled with SGI's third-generation NUMAflex(TM) architecture. SGI's NUMAflex global shared-memory architecture enables scientists to hold entire data sets in memory, allowing for faster and more interactive data analysis, and resulting in more incisive conclusions. Altix systems also offer the unique ability to scale memory, CPUs and I/O bandwidth independently of one another, so systems can be configured and easily upgraded to meet any challenge. "It's exciting to see that the SGI Altix supercomputer now in place at the NSC in Linkoping will serve as a vital engine for innovation and discovery for researchers throughout Sweden," said Afshad Mistri, segment manager for life sciences, SGI. "NSC's important work on SGI Altix reveals what is possible with a high-throughput architecture powered by industry-standard, open source technologies. From the entry-level, low-cost Altix 330 servers up to the powerful Altix 3700 Bx2 supercomputer, SGI Altix scales not only to meet HPC needs today, but can address even more demanding requirements in the years to come."