Volume shipments of Cray SV1Mountain View 23 Sep 99Horst Simon, director of NERSC, said: “This Cray SV1 will aid our scientific users in their work on some of the most important problems facing science today. Our vector supercomputers are a vital part of our computing arsenal. For certain applications, the unique processing power and high bandwidth brought by vector machines such as the Cray SV1 are absolutely essential.” Stan Burt, Ph.D., director of the National Cancer Institute’s Advanced Biomedical Computing Center, called the Institute’s new Cray SV1 “a significant resource for the entire biological research community. It is a powerful machine, with lots of memory, and users will find it to be a valuable tool in sorting out the functions of the genes and proteins involved in their diseases of interest.” Steve Oberlin, vice president of Cray Business Unit, SGI, said customer deliveries of the Cray SV1 system mark achievement of a key milestone on SGI’s high-performance computing roadmap. “With initial deliveries of Cray SV1 systems, we moved considerably closer to a single high-performance architecture unifying the best of parallel vector and scalable parallel technologies,” stated Oberlin. Cray SV1 supercomputers are fourth-generation CMOS vector systems designed to handle a broad range of vector applications. Each Cray SV1 node features two types of processors: an ultra-high-performance 4.8 gigaflops Multi-Streaming Processor (MSP) that handles computation-intensive applications and a standard processor with 1.2 gigaflops (A gigaflop is one billion calculations per second.) of peak performance for less-demanding applications. This combination, which enables the world’s first adjustable-size vector processors, lets users match their requirements with the system’s resources and allows Cray SV1 systems to handle varied problems and workloads efficiently. A fully configured Cray SV1 node contains six MSPs and eight standard processors. Processors are configured in a symmetric multiprocessing architecture similar to that used in Cray T90 and Cray J90 series supercomputers. Cray SV1 systems are scalable up to 32 nodes and one teraflops (trillion calculations per second) of peak performance. This scalable vector capability means most jobs can be run on a single Cray SV1 node, providing maximum ease of use, while a large Cray SV1 configuration can run multiple vector applications at once. Very large jobs can be run across multiple nodes using message-passing programming models. Cray SV1 processor performance is further enhanced by the system’s incorporation of the world’s first vector cache memory. “The combination of vector cache memory with our proven Autotasking technology delivers up to 25.6 gigabytes per second of memory bandwidth for each Multi-Streaming Processor,” said Dungworth. Autotasking requires minimal user intervention while permitting customers to run their vectorized, third-party application codes immediately. “With autotasking, our customers can easily optimize their application codes for peak performance at a high degree of granularity and in a process required only once for each code,” said Dungworth.
|
|