From chemistry-request@ccl.net Fri Feb 14 00:13:00 1992
Date: Thu, 13 Feb 92 23:37:01 -0500
From: fredvc@esvax.dnet.dupont.com
To: %chem@ccl.net
Subject: OS-s versus Compilers
Status: R

A cautionary comment on this response.

>>Regarding Doug Smith concerns on the use of VMS I would like
>>to add that we have a group of VAX 8600 where programs
>>run about 200 times slower than on cray ymp. We understood they
>>should run 30-50 times slower considering the speed and vector
>>capabilities of the ymp. We suspect it is a problem of VMS 
>>I would appreciate any comments on the conversion to ULTRIX

(1) 8600's are not very fast VAXen.  We found that speed doubled on going from 
    a 8600 to an 8650, with more "juice" from an 8800 and a little more from
    a 6000-400.

(2) CRAY Y-MP speed comes from several places: (a) a fast clock; (b)
    direct access to memory, rather than cache architecture; (c) very good
    SCALAR optimization; (d) Vector operations as an integral part of the
    system, rather as an add-on, along with good compiler recognition of
    vectorization opportunities. 

(3) The scalar/vector stuff (2c, 2d) is realized via the COMPILER, rather than
    the operating system, per se.

There may be good reasons for converting from VMS to ULTRIX, but I would not
place performance high on that list.  Unless you can get a better compiler in 
the deal, I would expect things to be pretty much the same.

++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++
                           FREDERIC A. VAN-CATLEDGE

Scientific Computing Division         ||   Office: (302) 695-1187
Central Research & Development Dept.  ||      FAX: (302) 695-9658
The Du Pont company                   ||
P. O. Box 80320                       || Internet: fredvc@esvax.dnet.dupont.com
Wiilmington DE 19880-0320             || 
++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++++