Abstract :
A description is given of the ES-1 supercomputer, which has a peak performance of 1.6 GFLOPS. The ES-1 is the first scientific computing system to achieve such performance using a moderately parallel architecture. Built using advanced silicon compilers and CMOS, the ES-1 is a multiprocessor with high-bandwidth interconnect and exceptional numerical performance. Unix C and Fortran 77 (with VAX extensions) are all supported, allowing easier porting of existing software in an environment that is familiar to most program developers. To facilitate multiuser access, the ES-1 supports network integration standards, such as Ethernet, TCP/IP, NFS, and X-Window systems. The ES-1 technical objectives, system architecture, and implementation are reviewed. Although the ES-1 is not a vector processor, its pipelined architecture and optimizing compilers allow it to process classical linear algebra problems efficiently. With extremely data-intensive programs the ES-1 delivers around 90% of its peak performance to a single user.<>
Keywords :
parallel architectures; parallel machines; 1.6 GFLOPS; CMOS; ES-1 supercomputer; Ethernet; Fortran 77; NFS; TCP/IP; Unix C; VAX extensions; X-Window systems; high-bandwidth interconnect; multiprocessor; multiuser access; network integration standards; numerical performance; optimizing compilers; parallel architecture; pipelined architecture; porting; scientific computing system; silicon compilers; supercomputing architecture; Computer architecture; Ethernet networks; Linear algebra; Optimizing compilers; Parallel architectures; Scientific computing; Silicon compiler; Supercomputers; TCPIP; Vector processors;
