SYSTEMS

SGI Altix 4700 Allows More Terabytes of SGI Global Shared Memory on Fewer Processors: Silicon Graphics, Inc. (SGI) global shared memory architecture delivers large shared memory systems, larger than any other vendor. With the SGI Altix 4700 system, SGI provides a unique large shared memory advantage per core: for example, 2 terabytes (TB) of main system memory can run on only 8 processor cores. SGI customers, many with up to 4 terabytes (TB) of global shared memory, experience world-class performance and a user-friendly environment that allows scientists to focus on their research, thus enabling breakthroughs to come faster and more frequently. In computer terms, "global shared memory" means that a system's main memory can be accessed from all of its processors, as opposed to distributed cluster memory (in which the four to eight processor cores on a single node can only access the memory, 16GB to 64GB of memory, on that node). A system that supports global shared memory provides a number of significant performance, productivity, and system administration benefits: -- Much larger and more detailed models of physical systems can be entirely memory resident, eliminating I/O and making processing more efficient. Consider, for example, modeling not only the turbine blades of an aircraft engine, but the complete engine. -- Applications that do random I/O on large data sets can experience up to a 1,000X performance increase by bringing the entire dataset into main memory. -- Decreased time to insight since applications can be rapidly prototyped with their full memory requirements and incrementally optimized using threads, message passing, partitioned global address space languages like Unified Parallel-C, or mixed programming paradigms. Around the world, SGI Altix 4700 systems are equipped with many terabytes of GSM. In the 1TB to 4TB system main memory range, SGI Altix system users include many hospital and biomedical research centers, universities, weather services and automotive engineering design. For example: -- The Pittsburgh Supercomputing Center acquired two SGI Altix 4700 systems earlier this year. The larger system (768 processors, 1.5TB of shared memory) is integrated into the TeraGrid, the National Science Foundation (NSF) program of comprehensive cyberinfrastructure, substantially increasing the shared memory capability available through NSF for U.S. science and engineering research. -- McLaren Racing is one of the most successful teams in the history of Formula 1, having won more than 150 Grands Prix, eight constructors' world titles, and 12 drivers world titles including the 2008 championship won by Lewis Hamilton. An SGI compute, storage and visualization customer since 2005, Vodafone McLaren Mercedes recently added an SGI Altix system with 1.3TB shared memory and 256 processors. The new system will run computational fluid dynamics (CFD) to simulate airflow, minimize drag and maximize top-end speed for race car design. -- The China National Satellite Meteorological Center (NSMC) recently deployed China's largest shared-memory computer -- and the fourth most powerful computer in the country -- to ensure the successful launch of China's new polar-orbiting meteorological satellite, FENGYUN-3. The SGI Altix system, with 1,280 processor cores and 4TB of shared memory, was initially used to fine-tune the software scientists and engineers at NSMC developed to effectively ingest, process and distribute data sent by the new satellite. It is now ready to handle the NSMC's remote sensing data preprocessing, replay data processing, simulation tests, Archival and Retrieval Service System (ARSS) database, and new application development requirements. With the SGI NUMAflex architecture, the SGI Altix 4700 system couples the large physical address space of the Intel Itanium 2 processor with a system interconnect capable of distributing that address space seamlessly across hundreds or even many thousands of nodes. "We see the capability of multi-terabyte large shared memory in the SGI Altix 4700 system as something that's highly differentiated and very interesting in the industry," said Bill Mannel, SGI director of server product management. "The growing number of customers selecting systems of this size speaks to the unique large shared memory advantage per core: you don't need vast numbers of CPUs to achieve large memory systems."