ENGINEERING

The U.S. Army's military ground vehicles of the future are being designed and tested with the help of SGI(R) technology. Silicon Graphics today announced that the Army's Tank Automotive Research, Development and Engineering Center (TARDEC) has upgraded its SGI HPC computation and visualization infrastructure with the addition of a new Silicon Graphics(R) Onyx4(TM) UltimateVision(TM) supercomputer and other SGI systems. SGI visualization and computational technology is instrumental to TARDEC's mission of designing the Army's next generation of lighter, more agile tanks, trucks, personnel carriers and other vehicles, and testing them in simulated, yet highly realistic, battlefield conditions. TARDEC's goal: deliver vehicles that support new strategic military objectives, while also identifying and eliminating design and performance problems long before the vehicles roll off the assembly line -- or onto the battlefield. TARDEC conducts real time "soldier-in-the-loop" (i.e., a driver operating the proposed vehicle design in virtual battlefield simulators) performance modeling and simulation, vehicle design analysis and optimization. Vehicle design characteristics, such as center of gravity, weight, spring rates, and inertial properties, are fed into simulation software running on the Onyx4 supercomputer. The vehicles are then run through their paces in virtual battlefield simulators and in tandem with manned motion base simulators with the Onyx4 system showing how the vehicles handle difficult terrains, high speeds, sharp turns and other stressful conditions in real time. Groups of TARDEC engineers and Army officers work collaboratively to observe and test vehicles' performances, share feedback and refine designs. Delivering 40 times the power at a fraction of the cost of previous systems, Onyx4 provides TARDEC a cost-effective, yet highly powerful solution. Onyx4 combines the power of SGI's unique, high-bandwidth, scalable, shared- memory architecture with best-of-breed, industry-standard graphics components. The Onyx4 UltimateVision system utilizes up to 32 tightly coupled best-of- class graphics processors capable of focusing on a single problem. The system enables visualization of larger data sets with greater complexity at higher interactivity, empowering individuals and improving team productivity. In addition to the Onyx4 UltimateVision system, TARDEC also installed a 24-processor SGI(R) Onyx(R) 3000 for visualization applications, a 32- processor SGI(R) Onyx(R) 350 for batch-oriented R&D computations, a Linux(R) OS-based SGI(R) Altix(TM) 3000 series server with eight Intel(R) Itanium(R) 2 processors for special requirements (such as frame rendering and real-time modeling and simulation algorithm development) and doubled its storage capacity to 5.6TB with an SGI(R) InfiniteStorage TP9100 Fibre Channel RAID array. The systems for TARDEC were installed in October of this year. The transaction also includes services over a three-year period. These additions represent an ongoing modernization effort conducted by TARDEC's National Automotive Center (NAC), which provides high-performance computational and visualization resources and expertise to TARDEC and partners. TARDEC's HPC architecture has evolved from its original utilization of a Cray-2(TM) supercomputer and other platforms from the early 1980s until the mid 1990s, when the organization began to adopt SGI HPC technology. Onyx4 UltimateVision solutions are based on the SGI NUMAflex(TM) architecture, which provides independent scaling for all of the major system components including the processors, memory, storage, I/O, and graphics. As components are added, Onyx4 automatically scales bandwidth, maintaining a balanced system and transparently maximizing application throughput and performance for any size problem. Distinctive from older computer architectures and clusters where every processor has its own associated memory, by not having to move data from one memory location to another, resulting in faster processing and easier programming. Simply put, utilizing the NUMAflex share memory model saves users tremendous time and effort.