New super-computing facility in RIKEN
- Institute of Physical and Chemical Research (RIKEN), Wako-shi (Japan)
A new superconductor, Fujitsu VPP500/28, was installed in the Institute of Physical and Chemical Research (RIKEN) at the end of March, 1994. It consists of 28 processing elements (PE`s) connected by a high-speed crossbar switch. The switch is a combination of GaAs and ECL circuitry with peak band width of 800 Mbyte per second. Each PE consists of a GaAs/ECL vector processor with 1.6 Gflops peak speed and 256 Mbyte SRAM local memory. In addition, there are 8 GByte DRAM space, two 100 Gbyte RAID disks and a 10 TByte archive based on SONY File Bank system. The author ran three major benchmarks on this machine: modified LINPACK, lattice QCD and FFT. In the modified LINPACK benchmark, a sustained speed of about 28 Gflops is achieved, by removing the restriction on the size of the matrices. In the lattice QCD benchmark, a sustained speed of about 30 Gflops is achieved for inverting staggered fermion propagation matrix on a 32{sup 4} lattice. In the FFT benchmark, real data of 32, 128, 512, and 2048 MByte are Fourier-transformed. The sustained speed for each is respectively 21, 21, 20, and 19 Gflops. The numbers are obtained after only a few weeks of coding efforts and can be improved further.
- Research Organization:
- Lawrence Berkeley National Lab. (LBNL), Berkeley, CA (United States)
- OSTI ID:
- 54330
- Report Number(s):
- LBL-35822; CONF-940492-; ON: DE95004442; TRN: 95:012530
- Resource Relation:
- Conference: Meeting on computing in high-energy physics, San Francisco, CA (United States), 21-27 Apr 1994; Other Information: PBD: [1994]; Related Information: Is Part Of Proceedings of the conference on computing in high energy physics `94; Loken, S.C. [ed.]; PB: 532 p.
- Country of Publication:
- United States
- Language:
- English
Similar Records
A high performance linear equation solver on the VPP500 parallel supercomputer
ID-1 mass storage system for mainframe by using FDDI network