Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Path: utzoo!mnetor!seismo!rutgers!sri-spam!ames!ucbcad!ucbvax!edison.ge.COM!uucp
From: uucp@edison.ge.COM (UNIX-to-UNIX Copy)
Newsgroups: comp.os.vms
Subject: Submission for mod-computers-vax
Message-ID: <8704280633.AA19865@edison.GE.COM>
Date: Tue, 28-Apr-87 02:33:12 EDT
Article-I.D.: edison.8704280633.AA19865
Posted: Tue Apr 28 02:33:12 1987
Date-Received: Sat, 2-May-87 03:08:18 EDT
Sender: daemon@ucbvax.BERKELEY.EDU
Distribution: world
Organization: The ARPA Internet
Lines: 62

Path: edison!uvacs!virginia!umd5!brl-adm!seismo!esosun!ucsdhub!sdcsvax!ucbvax!UKANVAX.BITNET!KRICE
From: KRICE@UKANVAX.BITNET
Newsgroups: mod.computers.vax
Subject: RE: Megaflop rating on the VAX 8600
Message-ID: <8704261101.AA02707@ucbvax.Berkeley.EDU>
Date: 25 Apr 87 04:34:00 GMT
Sender: daemon@ucbvax.BERKELEY.EDU
Distribution: world
Organization: The ARPA Internet
Lines: 51

Hello out there in NetLand:

Someone using the account RMB059@MILPL3.JPL.NASA.GOV asked:

>What's a good number for the Megaflop rating for an 8600?

According to a study named "Performance of Various Computers Using Standard
Linear Equations Software in a FORTRAN Environment" by: Jack J. Dongarra,
Argonne Natl. Lab., Argonne, Ill. 60439 U.S.A., Dated: Sept. 24, 1986:

The study used four different determinations of megaflop ratings.  All
comparisons were done using a benchmark program called LINPACK [ownership and
specifics unknown.  It is only said that it is a system of linear equations.
Other specifics unknown, sorry.], written in FORTRAN.  All use VMS V4.1 and
FORTRAN V4.2. They gave the following results:

                              Ratio: vs. Cray-1s       Megaflops
                              ******************       *********

1> (Full Precision, all FORTRAN):    1 / 25              0.49

2> (Full Precision, coded BLAS):     1 / 19              0.66

3> (Half Precision, all FORTRAN):    1 / 14              0.88

4> (Half Precision, coded BLAS):     1 / 9.8             1.3

No, BLAS isn't what happens on Monday mornings.  The author says that by using
"coded BLAS" they rewrote the part of the code that does most of the floating
point maneuvering in assembly.  The report was supported by the DOE, so the
U.S. Gov. Superintendent of Documents probably has a copy.

I should probably include a disclaimer at this point that I have no association
with anyone (except my family, and even they deny me sometimes), so don't
blame/attribute to/commend/reproach/bespatter/condemn/denounce/censure/
reprehend/accuse/admonish/rebuke/castigate/trounce/chastise/discombobulate/
impeach/challenge or injure/pulverize/overmedicate/wound/hurt/deface/mar/impair
me for anything said above or below (next to, etc., you fill it in). I have a
photocopied version of this report (oops, hope there's no copyright
infringement here... oh well), so contact the writer (above) please, not me.
Yokee-doeekee, that's all I know, folks.

                                  Kevin Rice
                        Student, University of Kansas
                           Lawrence, Kansas 66045

                           BITNET:  KRICE@UKANVAX
                           CSNET:   KRICE@KUHUB.UKANS
                           MaBell:  (913) 864-2919
   Ocean:   Drop it in, it just might get here (This *IS* Kansas, the land
                   of Ahhs, seaport to the entire midwest).