Nobuo Uchida

Fujitsu VP2000 series

A description is given of the Fujitsu VP2000 series, a new-generation supercomputer supporting full compatibility with previous VP and VP-E series, and with much higher performance. The VP2000 series has a performance range of between 0.5 and 4 GFLOPS. It features a dual scalar processor architecture, expanded main storage, system storage, and connectivity to UNIX environments. The series offers a wide range of vector performance, with the most powerful model being eight times faster than the entry-level model. Whereas Model 10 has a uniprocessor architecture, Model 20 is a dual scalar processor system in which two scalar units can share one vector unit. The VP2000 series has a total of eight models that are field upgradable. The new features in architecture, LSI technology, and software are discussed.<<ETX>>

FUJITSU VP2000 SERIES SUPERCOMPUTERS

Fujitsu VP2000 series Supercomputer is a pipeline supercomputer series developed by Fujitsu. The peak performance of this series ranges from 375 MFLOPS of the VP2100/10S to 5 GFLOPS of the VP2600/10. Operations are mainly performed by using a pipeline which is called “Universal Pipeline”. This pipeline provides a capability of one of the following operations such as a multiplication, an addition, a multiplication-addition or a logic operation at one time. In addition, in case that two pipelines are used, they can perform such operations concurrently. The VP2200, the VP2400 and the VP2600 have two pipelines and both pipelines are used for multiplication-addition concurrently to attain the peak performance. FORTRAN77 EX/VP, which is developed to utilize this hardware, makes it possible to schedule these two pipelines effectively. Using the VP2600/10, the performances of LINPACK problem for a matrix of order 100 and for a matrix of order 1000 are 249 MFLOPS and 4009 MFLOPS respectively.