S group;
A while ago I queried the group as to whether people could make good use of
a massive speed increase in their S+ code. This was so I could determine if
it was worthwhile to attempt a naive port of some S+ Fortran routines to run
on Cray's UNICOS system. It's very similar to UNIX but there are some very
ugly contraints on the way they optimize memory. Double-precision numbers
are difficult to use because they are a drag on optimizing formulas for
speed. Regardless, I am still working on a port of a Fortran ANOVA routine
(standard code from S+) to UNICOS and will present the comparison values to
the group as soon as I am able. It's been a little challenging working with
two companies who are quite busy with other projects but both Cray and
Insightful have been very helpful. I believe an increase of 1000x to 10000x
over a 400MHz Sun Machine is possible for some types of number crunching
(based on raw performance numbers).
Phillip
|