In spite of the progress made recently with parallel machines and the
variety of applications which have been adapted for parallel techniques, there
remain a variety of numerically intensive applications requiring a fast single
processor. A vector processor is just such a fast single stream machine with
additional hardware to accelerate repeated similar calculations as might be
applied to arrays of data.  The University is fortunate in having a Gould NP1
machine on loan for a joint communications project and this is providing
valuable experience in this area. It is possible that the loan will be come
permanent if the University can raise fund to cover the maintenance.

     The magnitude of acceleration provided by a Vector processor is limited; a
factor of four or five seems to be the limit and the average increase in
performance of a lengthy and varied calculation is typically two to two and half
times. However Vector processors have one priceless advantage in that
reprogramming is not normally needed. The minor rearrangement of the (normally
Fortran) source is within the capabilities of current optimizing compiler
technology and the enhanced performance can usually be gained simply by
recompilation. Further accessible enhancement will come when certain scientific
subroutines like the NAG Library are available in vector form. For this reason
Vector processors are likely to have a permanent place in research orientated
computing