Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.2 9/18/84; site petrus.UUCP
Path: utzoo!watmath!clyde!burl!ulysses!allegra!bellcore!petrus!hammond
From: hammond@petrus.UUCP
Newsgroups: net.unix-wizards
Subject: 11/785 FPA problems
Message-ID: <329@petrus.UUCP>
Date: Tue, 23-Apr-85 07:58:45 EST
Article-I.D.: petrus.329
Posted: Tue Apr 23 07:58:45 1985
Date-Received: Wed, 24-Apr-85 03:54:16 EST
Distribution: net
Organization: Bell Communications Research, Inc
Lines: 11

While doing benchmarks on floating point performance, I noticed that our
11/785 with FPA took several times as long as an 11/750 w/FPA to do
double multiplies.  Our DEC salesman claims that the 11/785 FPA has
a microcode bug in the multiply and that the machine is wired to ignore
the FPA for multiply.  I think he claimed that there should be an ECO
out for this (new microcode).  If you have an 11/785 with FPA, check
by multiplying two non zero numbers (try 1234567.8 *  10) a million
times.  If you have the bug it willl take about 20 seconds (i.e. 20
micro seconds per loop), with a fixed FPA it should take 5 to 10 seconds
(our 11/750 takes 7.7).
Rich Hammond	{allegra | decvax | ucbvax}!bellcore!hammond