MOUNTAIN VIEW, Calif. -- SGI today announced that Germany's Waterway Engineering and Research Institute (BAW) heavily invested in SGI(R) technology for high-performance computing (HPC) to meet challenges in future numerical simulation projects. BAW's coastal division in Hamburg, which designs, monitors, analyzes and repairs all waterways and tidal flood zones in the country's coastal areas, has installed two of the new SGI(R) Origin(R) 3000 servers. The two supercomputers harness the power of 288 MIPS(R) processors and run the high-performance SGI(R) IRIX(R) operating system. The acquisition worth approximately 3 million euro (roughly $3.2 million U.S.) will not only greatly increase computing power for performance-hungry users, but will also establish an HPC environment that offers highly flexible usability and uniquely enhances supercomputing productivity. The new SGI platform replaces BAW's former vector-based supercomputer system, Cray SV1(TM), and won the business over other major vendors' competitive system offerings by delivering convincing benchmark performance and the best price/performance value. Using computer modeling and numerical simulation techniques, BAW achieves an increasingly precise understanding of the complex physical processes that govern fluid dynamics and suspended material transport phenomena within tides-influenced coastal waterway systems like the big river mouths of Elbe, Weser and Ems. "This installation further validates that technical computing customers such as BAW require more than just raw CPU megahertz speed improvements. They need balanced capacity and capability computing systems for mission-critical problems. They need supercomputers that are designed for the data-intensive, complex technical computing environments in which they work, and the SGI Origin 3000 family was designed for just this purpose," stated Jan Silverman, senior vice president of marketing at SGI. BAW's main job is to analyze the consequences of planned constructive measures. Such measures could include deepening a shipping channel, optimally designing and operating a barrier system against dangerous storm floods, or performing other technical operations that may critically impact the dynamic behavior of the river estuary systems. Higher Throughput, Better Physics and Better Stochastic Models The new SGI systems will allow the BAW team to run a greater number of simulation projects and get more precise answers to its many technical, economical and ecological problems and questions. With the SGI Origin 3000 systems, BAW users will be able to refine its modeling resolutions, to introduce more adequate physics, to run simulation models in 3D instead of 2D-and, very importantly, to better take into account the impact of weather and associated stochastic variations. The two SGI(R) Origin(R) family systems are configured as a production platform with 256 processors and a development platform with 32 processors, which, if needed, may also be used as a second or redundant production machine. In both systems, all processors are able to commonly address the complete installed memory area -- and thus can jointly and efficiently process jobs with different problem sizes-all within a flexibly usable shared-memory parallel computing environment. For each processor, the BAW installed half a gigabyte of memory. SGI Systems Score High in Benchmarking, Are Quickly Ready for Production The systems, ordered in November 2002 and installed in December, passed the customer's approval check and came online in January 2003. By mid-February, they were ready for full production use. The SGI platform prevailed after passing an intensive selection and benchmarking process that included all major HPC technology vendors across Europe. BAW's new system had to meet a number of clearly defined requirements that were mainly performance-centric but also included criteria like how well the new systems could integrate into the existing environment. Benchmarking tests included two of the approximately 50 computational fluid dynamics (CFD) application software codes the BAW team uses. Because the software packages were available in source code, vendors could port to their platforms and also optimize the code to take complete advantage of the specific architectural benefits. The benchmark tests measured how the systems processed and solved problems during a variety of jobs. Dr. Harro Heyer, head of simulation-based projects at the BAW's coastal division in Hamburg stated: "Under all offerings, the SGI Origin system proved to have the best price/performance value. And it's really exciting to see: Though the MIPS processor is not clocked as high as those of the competitors, SGI's machine deliver higher application performance in a real-world production environment. Here is a system architecture with great scalability and bandwidth features, software and hardware are playing efficiently together, and the SGI team succeeded in delivering the best optimized code." "We had also been impressed with how quickly the SGI team could install and integrate the machine into our existing environment, so that we have been able to get productive in such short a time frame," added Dr. Heyer. Migrating from Vector to Microprocessors The institute is also executing a migration from vector-based supercomputing to HPC within a massive parallel processing environment. "Going from vector to parallel, it is very important for us to be able to do shared-memory computing," said Dr. Heyer. "A system that allows us to have 32, 64 or, if required, even 256 processors working on a single problem -- and to have them all commonly address a single large memory space -- will provide substantial benefits." Heyer finds the SGI architecture for shared-memory computing to be ideally suited for optimizing BAW's application software performance. The CFD codes have to be run with unstructured meshing networks to account for the intrinsic aerial resolution variations. Some topographies, such as the environments surrounding construction sites, require high-detail model treatment; other areas may become completely dry within a tidal simulation cycle and not require as high degree of detail. "Unlike message-passing computing, shared-memory computing enables us to load-balance more effectively and to distribute the number-crunching tasks more efficiently to the many processors. Furthermore, we are able to adapt such a machine more easily and more effectively to the different throughput requirements of the various job profiles and job mixes we have in Hamburg," added Heyer.