Relay-Version: version B 2.10 5/3/83; site utzoo.UUCP
Posting-Version: version B 2.10.3 4.3bsd-beta 6/6/85; site ucbvax.BERKELEY.EDU
Path: utzoo!decvax!decwrl!ucbvax!apollo
From: KROWITZ@MIT-MARIE.ARPA
Newsgroups: mod.computers.apollo
Subject: (none)
Message-ID: <8601131532.AA02839@Yale-Bulldog.YALE.ARPA>
Date: Mon, 13-Jan-86 10:52:33 EST
Article-I.D.: Yale-Bul.8601131532.AA02839
Posted: Mon Jan 13 10:52:33 1986
Date-Received: Tue, 14-Jan-86 04:17:46 EST
Sender: daemon@ucbvax.BERKELEY.EDU
Organization: The ARPA Internet
Lines: 44
Approved: apollo@yale-comix.arpa

I've been away for awhile and haven't been reading my mail until recently.
Thanks for the Fortran speed comparisons between the DN330 and the Sun 3.
In the meantime a 330 upgrade board showed up on the doorstep of another
group in our ringnet, and I've taken the liberty of running a Whetstone
program on some of our nodes. For the record:

machine/OS			single precision	double precision
----------			----------------	----------------

DN320/SR9.0.020			308			190
DN330/SR9.0.020			714			658
DN460/SR9.0.020			689			425
DN660/SR9.0.020			723			446
VAX 11/782/VMS 4.2		1230			794
VAX 11/780/BSD 4.2 (UNIX F77)	399			364
Alliant FX/1			3953			3289

Note that an 11/782 is a dual -cpu 11/780 system in a master/slave
configuration. It is the same speed as an 11/780 for a single job.
The Alliant FX/1 is being marketed by Apollo along with its big
brother the FX/8. We have a very early production model
bought directly from Alliant with Rev. 1 processor boards. Rev. 2
boards are claimed to be 20% faster. For comparison, the numbers for
a Sun 3 (from David Hough's message) are:

Sun 3				860			790

Which is about 20% faster than the DN330. Note that the DN330 is
faster than any of the other Apollo machines for double precision
arithmetic (as is the Sun 3) -- an important feature for us
engineering/scientific number grinders. The DN330 (2MB) is $15,900
(list price). Anyone out there have the price on a comparable Sun 3
(ie. 2MB diskless node without compiler licences)? As it was
noted earlier, the DN660 (which used to be Apollo's fastest cpu)
is a *lot* more expensive at a list price of  $52,500 (but then
it's an older technology machine, 4MB, and a color screen and graphics
processor so it's bound to be more expensive).

As a parting note: Anybody got a copy of the Linpack
benchmarks (or the Livermore loops) that they can spare?

					-- David Krowitz
				         ( DAVID@MIT-MC.ARPA or
					   KROWITZ@MIT-MARIE@MIT-MC.ARPA )