| The AlphaServer SC series combines the awesome computing power and memory bandwidth of Alpha processors with the performance, scalability, and ease of use of the SC interconnect, SC single-system-image software, and SC support. This series has been and continues to be selected by the leading research institutions to solve their most challenging, scalable simulation and engineering tasks. AlphaServer SC systems are the solution of choice for maximum sustained processor performance, scalability, interprocessor communication, single-system-image ease of use, and high-bandwidth, global I/O. AlphaServer SC features The AlphaServer SC series uses standard volume components with the SC interconnect and SC software to build terascale supercomputers that include: - Scalability from 25 GFLOPS to 10s of TFLOPS. - AlphaServer SMPs, including AlphaServer ES45 and AlphaServer ES40 servers, with high sustained performance delivered by up to 8 GB/s of memory bandwidth per server. - AlphaServer SC interconnect, with over 500 MB/s of interprocessor message-passing bandwidth per node and 32 GB/s of bisection bandwidth. - AlphaServer SC system software, with single-system-image installation, maintenance, resource management, and scalable I/O. For example, systems have been designed to checkpoint a distributed 1 TB image to disk in under 15 seconds. - AlphaServer SC filesystems, providing scalable-shared bandwidth from each scalable node to common-shared, high-bandwidth filesystems. - AlphaServer SC resource management, including Platform Computing's LSFTM, to optimize and control the flow of work for maximum throughput and resource utilization. - Parallel application development environment and tools, for maximum delivered performance with a minimum of porting and development effort. Above all, AlphaServer SC supercomputers are designed to grow and grow to terascale systems and beyond. The AlphaServer SC45 scales up to 4096 processors (10 TFLOPS), with larger (10s of TFLOPS) systems available. Deliveries of AlphaServer SC45 systems in 2001 included several terascale systems, including the Terascale Computing System (TCS) at Pittsburgh Supercomputer Center (PSC), the largest civilian supercomputer ever delivered. This system, with 6 TFLOPS of peak performance across 3040 Alpha EV68 processors, has 8.9 times the peak performance of the original AlphaServer SC40 systems. The AlphaServer SC interconnect delivers the best low-latency, scalable bandwidth and I/O available, enabling a wide range of applications to scale efficiently across tens, hundreds, and even thousands of processors. The AlphaServer SC45 system delivers over 500 MB/s of message-passing bandwidth per node with less than 5 microseconds of MPI latency. This unique combination of high bandwidth, low latency, and high scalability minimizes the overhead of interprocessor communication, allowing applications to scale efficiently to deliver higher performance than on any other scalable system. The AlphaServer SC interconnect, in combination with SC filesystem software, also delivers high-bandwidth, scalable I/O from the distributed-scalable processors to common-shared filesystems. The AlphaServer SC Series is designed for performance growth, following HP's strategic roadmaps that cover both processors and servers. AlphaServer systems will continue to deliver balanced memory bandwidth for years to come. Next-generation AlphaServer memory interconnects are based on Alpha EV7 processor technology, with many times the peak memory bandwidth per processor. Following that generation, Alpha technology and resources will accelerate and enhance the Intel® ItaniumTM processor family, for continuing sustained-performance leadership. Leading research institutions consistently choose AlphaServer SC systems for their midrange-to-terascale High Performance Technical Computing (HPTC) needs. AlphaServer SC systems include some of the largest terascale supercomputers in the world. Leading AlphaServer SC sites include Pittsburgh Supercomputer Center (PSC), Los Alamos National Laboratory (LANL), the French Atomic Commission (CEA), Japan Atomic Energy Research Institute (JAERI), Australian Partnership for Advanced Computing (APAC), Victorian Partnership for Advanced Computing (VPAC) and the Lawrence Livermore National Laboratory (LLNL). |
 |