Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.1 6/24/83; site mit-eddie.UUCP
Path: utzoo!watmath!clyde!cbosgd!ihnp4!mit-eddie!smh
From: smh@mit-eddie.UUCP (Steven M. Haflich)
Newsgroups: net.unix-wizards
Subject: Floating point benchmarks for 11/73?
Message-ID: <3201@mit-eddie.UUCP>
Date: Wed, 28-Nov-84 07:07:56 EST
Article-I.D.: mit-eddi.3201
Posted: Wed Nov 28 07:07:56 1984
Date-Received: Thu, 29-Nov-84 05:54:10 EST
Distribution: net.unix-wizards
Organization: MIT, Cambridge, MA
Lines: 20

Now that folks have begun accumulating real experience with the 11/73
I'd like to hear if anyone can provide any benchmarks or even fuzzy
impressions of its floating point speed.  With good reason, DEC is
unwilling to release "instruction-timing" numbers.  These are usually
meaningless anyway (compare the 11/45!) because the overhead times for
communication between the FPP and the rest of a system can be
significant and very difficult to calculate.

My group is considering inexpensive networked 11/73 engines as
downloaded processors for some venerable production signal processing
code still running on an expensive PDP11/45.  Fixed-point comparisons
are easy to come by, but the code in question makes *extensive* use of
floating instructions.  Typical code has up to a 50% mix of floating op
codes, frequently hand intermixed with non-floating ops to take
advantage of execution overlap on the 11/45.

I'll summarize to the net.

Steve Hafich
{ihnp4, decvax!genrad}!mit-eddie!smh