|
NEC Corporation thanks you for requesting information about our high performance computing products. This email response is an entry into the world of NEC, which can be more fully explored by visiting our web sites or contacting our staff. Our primary web sites include:
Division Home PageThe content of this email response includes:
SX Series Systems -- Supercomputers for All Computing EnvironmentsSX-6 Series systems are today the leading shared memory parallel vector processor systems available. They lead the industry in both power per processor and memory bandwidth. Time again they have proven to be a uniquely cost effective applications solution environment for users needing true supercomputer class performance. The SX-6 breaks new ground with it's new single chip vector processor. This means nothing less than that the SX-6 vector supercomputer is a microprocessor-based system, thereby entering a previously unreachable price/performance level. The SX-6 Series product range provides an unequalled balance of processing power, memory capacity, memory bandwidth, and I/O bandwidth to meet the most demanding turnaround and throughput requirements. SX-6 Series supercomputers are the systems of choice world-wide for both mission critical capability computing as well as large-scale capacity requirements. The balanced NEC SX architecture provides the highest levels of performance for legacy and newly developed applications, and is most importantly easy to program. During the previous decade there has been a trend away from vector parallel systems such as the SX Series, toward systems built on commodity technology. However, a number of critical applications areas have not been well supported by them, including:
Because of very high software development investment required for the development and certification of scalable applications, or because of inherent limited parallelism within applications, systems with shared memory and powerful individual processors have continued to be the platforms of choice for these solutions. Shared memory supercomputers with high memory bandwidth, such as the SX-6 Series, are again recognized as necessary to achieve the levels of performance required for advancing the respective sciences. Since the introduction of the all CMOS NEC SX-4 Series in 1994, the economics of real parallel vector supercomputers have been highly competitive with systems based on collections of workstations (COWS). It has also been experienced that for many applications a COWS-based system is not a substitute for a real parallel vector supercomputer. The introduction of the single-chip vector processor in the NEC SX-6 is another major step in the same direction. SX-6 Series models span a range from high end server class systems, to models targeting the highest performance mission critical requirements. They span a performance range from an 8 GFLOPS rated single processor through an 8 TFLOPS 1024 processor multinode system. The SX-6 application library includes all major commercially available and supported applications relevant to supercomputer class problems, as well as many public domain codes that have been ported and optimized. The SX-6 Series -- Capability Supercomputers for the 21st CenturyThe performance range of SX-6 Series models now spans from 8 GFLOPS to 8 TFLOPS through an upward compatible system family. The operating system, SUPER-UX, is a supercomputing enhanced UNIX and is a robust, full featured, evolutionary product that has been in widespread commercial use since 1990. SX-6 Series systems are offered as single node and multi node models. The single node frame is the basic building block for all models. An SX-6 single node can be configured with up to 8 processors. This provides a performance range up to 64 GFLOPS with a shared memory symmetric parallel vector architecture. Single node shared memory capacity ranges up to 64 GB with bandwidths up to 256 GB per second. SX Series Systems simply provide high performance solutions. SX-6 Series TechnologyThe SX-6 Series is designed with CMOS VLSI. This provides for a highly competitive price structure, low power consumption, and air-cooling. SX-6 Series ProcessorThe SX-6 processor is implemented on just one CMOS LSI, the SX-6 can therefore be called a microprocessor-based machine. The SX-6 operates at a cycle time of 2ns or a clock frequency of 500 MHZ, respectively. Each SX-6 Series processor is rated at 8 GFLOPS peak performance. The memory provides 32 GB per second bandwidth to each processor, enabling a high sustainable performance level. The SX-6 Series scalar unit can issue up to four instructions per clock. It features branch prediction and dynamic scheduling allowing out-of-order issue sequencing for vector, scalar, and memory reference instructions. The technologies adopted in the scalar unit rank among the most sophisticated available even for any leading scalar-only processor design. The SX-6 Series processor supports both 32 and 64 bit scalar and vector IEEE data and ads extended precision IEEE in the scalar unit. NEC also recognizes software investment, and as with previous SX generations, the SX-6 is binary compatible with IEEE mode SX-4 and SX-5 applications load modules. SX-6 Series MemorySX-6 Series models utilize 256 Mb DDR SDRAM (Double Data Rate Synchronous Dynamic Random Access memory) main memory parts. Single node shared memory can be configured up to 64 gigabytes. 32 GB per second of bandwidth is available to each processor, providing a full 256 GB per second bandwidth for a full single node configuration. SX-6 Series Multi Node SystemsSX-6M multi node models are constructed using single node frames as building blocks and the NEC IXS Internode Crossbar Switch (IXS) to interconnect multiple nodes. Multi node configurations utilize physically shared memory within each node, and distributed memory across nodes, comprising shared-NUMA memory architecture. The IXS is a full crossbar that includes hardware support for global addressing, global synchronization, and globally extensible instructions. The SUPER-UX operating system software supports multi node global addressing, however message passing programming models are used to take maximum advantage of the underlying hardware. The result is an integrated single system image system. Each SX-6M node has 8 GB per second bandwidth to the IXS. Latency due to transmission and switching time is consistent between nodes regardless of how many nodes are configured. The IXS provides up to 1 TB per second bisection bandwidth. SX-6M Multi Node Shared and Distributed MemoryAll SX-6 Series models feature uniform physically shared memory architecture at the single node frame level. The multi node environment is viewed as distributed NUMA memory even though it is globally addressable. Until an application requires more memory capacity than a single node frame can support, all of the advantages and simplicities of shared memory programming can be enjoyed. The complexities of distributed memory need not be addressed until the application requires more than 64 GB memory space, or more than 64 GFLOPS of compute power-the single node limits. Further, message passing programming models can be effectively executed on shared memory more efficiently than is possible on distributed memory. When the multi node environment is required for a large application, up to 8 TB aggregate memory capacity is available with up to 8 TFLOPS of compute power. The multi node programming model can be message passing or a combination of traditional multitasking and message passing. The result is high sustainable performance for applications based on either shared or distributed memory programming models. It also provides flexibility for developing distributed memory applications for SX-6 Multi node configurations without impacting legacy production applications designed strictly for shared memory. SX-6 Series I/O SubsystemThe Input/Output Subsystem is designed to provide I/O performance commensurate with the processing power available. FC-AL, FWD SCSI, Ultra SCSI, Ethernet, and FDDI connections are all directly supported by the SX-6 Series IO subsystem. Each SX-6 Single node can be configured with up to 8 gigabytes per second of I/O bandwidth. Multi node models can be configured with 8 GB per second per Single node frame, or up to 1 TB per second in a maximum configuration. SX-6 Series ModelsAn SX-6 Series single node full sized chassis holds up to 8 processors, 64 GB of main memory, and 4 IOPs. The full sized cabinet is called an "A Cabinet". The SX-6 Series half-sized chassis can hold up to 4 processors, 32 GB of main memory, and 2 IOPs. The half sized cabinet is called a "B Cabinet". SX-6 multi node models can be configured from 2 to 128 nodes; each node consisting of an SX-6 Series chassis. The multi node model utilizes the IXS and can be configured with up to 1024 processors delivering 8 TFLOPS peak performance with 8 TB of main memory having an aggregate maximum memory bandwidth of 32 TB per second. SX-6 Series NetworkingAll UNIX communications functions and protocols are supported over all common network media. The SUPER-UX Operating SystemSUPER-UX is a UNIX based operating system that has been enhanced with features required for supercomputing environments. It is reliable, efficient and provides the broad services and advanced features expected today. Among the operating system enhancements are an enhanced NQS batch subsystem, multilevel I/O caching; extensive parallel processing support including a multi node job class; versatile resource management; extended accounting, checkpoint/restart, and fully automatic operation facilities. The SX Series is intended for technical computing applications, and as such supports a comprehensive suite of compilers and development tools to aid in application development or porting. The compilers available for the SX-6 Series include Fortran90/SX, HPF/SX, and C++/SX. Full, automated support is provided for vector processing and shared memory parallel vector processing. All configurations of the SX Series support traditional multitasking (FORTRAN microtasking and C threads) as well as MPI message passing programming models and OpenMP. SUPER-UX provides a total system that is seamlessly integrated, reliable, full featured, and efficient in either single node or multi node configurations. PSUITE Integrated Software Development EnvironmentThe PSUITE environment integrates Fortran90/SX, C++/SX, data analysis, program tuning, debugging, source editing, and project management tools into a coherent Integrated Development Environment. PSUITE includes native and workstation based Fortran90 cross compilers and optimization tools. All the compilers provide automatic vectorization and parallelization functions. An optimized MPI message-passing library is supported and PVM is available. PSUITE includes tools to assist in writing, debugging, and tuning parallel programs, including those in the MPI message passing model. Company BackgroundNEC Corporation was founded in 1899 and has headquarters in Tokyo, Japan. We are one of the world's oldest and largest designers and manufacturers of supercomputers, computers, electron devices, and communications equipment. Our worldwide organizations employ approximately 150,000 people generating gross revenues of approximately $US 42 Billion annually (at Y126 = $US 1). We have 89 consolidated subsidiaries, 53 manufacturing plants and 440 sales offices in Japan. Worldwide we have another 39 consolidated subsidiaries and 49 majority owned subsidiaries having an additional 52 plants. Our strong commitment to North America is visible through a network of operations located coast to coast that employs 8,000 people. These operations include 8 manufacturing plants, and research and development activities in core technologies, as well as sales and support facilities for our diverse product range. In the United States of America, Canada and Mexiko our high performance computing products are sold exclusively by Cray Inc (see http://www.cray.com) under an OEM agreement. The NEC SX-6 is marketed as the Cray SX-6. Cray Inc also holds the rights to sell the Cray SX-6 in other parts of the world on a non-exclusive basis. In Europe our high performance products are represented by the NEC European Supercomputer Systems Operation (ESS). ESS maintains sales and support offices throughout Europe. We have 4 major manufacturing facilities in Europe, research and development facilities, and numerous sales and support offices employing about 4,000 people. In addition to NEC ESS and NEC Australia offices, specialist high performance computing representation is available at NEC offices in Singapore and Sao Paulo, Brasil, as well as throughout Japan. Our unique multinational operating model allows us to provide industry-leading response to our customers needs throughout the world. Our sales, support, and business operations are global in scope and provide unsurpassed opportunities to continue serving satisfied customers. As evidence of our capability, We were selected to develop the Japan Earth Simulator Program Ultracomputer. This will be a 40 TFLOPS parallel vector system having 4 TB of Main Memory. It will be completed by 2002 and will represent the most powerful general purpose computational resource ever built. It will primarily be used for complex modelling of the earth's climate, geology, for advanced studies in the socio-economic impacts of global warming, and for prediction and resolution of various natural disasters or environmental catastrophes. ContactsNEC's worldwide sales and support organization for high performance computer products include major units located in:
|