koko@uthub.UUCP (10/15/87)
Received: from jrn by mill.me.toronto.edu via UNIX id AA09433; Wed, 14 Oct 87 21:59:08 EDT
Date: Wed, 14 Oct 87 21:59:08 EDT
From: "John R. Nickerson" <jrn>
Message-Id: <8710150159.AA09433@mill.me.toronto.edu>
To: kokody2
Status: R
The inaugural presentation for the fall UTME seminar series
was held on Tuesday, October 6th./1987 in MC252 (Mechanical
Engineering) at 1300hrs. (1:00pm). The topic was on the
latest architectural advances for minisupercomputers.
The Multiflow Trace Series Computers:
Trace Scheduling and VLIW Architectures
By: Dr. Joseph Fisher, Vice-President and Founder
Multiflow Computer Incorporated
Josh made a lucid and convincing case for VLIW (Very Long
Instruction Word) computers. His presentation was
professional and polished with a plethora of benchmarks
being presented. The unique thing with the Trace series
computers was not the use of overlapped instructions per se
but rather the symbiosis between hardware and software
resulting in 7 possible instructions being executed at one
time. The basis for doing this is left to the compiler. The
machine has a very a slow clock (130 ns.) and consequently
it has tremendous growth potential based slowly on speeding
up the clock (within memory bandwidth limitations). After
the formal seminar several of us had a small private
discussion with Josh about future models and the limitations
of word width within the VLIW concept. The benchmarks for
the Trace 7 looks like:
Compiled Linepack (Full Precision, MFLOPS) 6.0
Whetstone (Double Precision KWHETS's) 12605
Livermore Loops (24 Kernels)
(Double Precision MFLOPS) 2.3
As a final note Josh was completely conversant with
virtually every detail of the machine and was able to answer
any question at the level it was asked at. The net result
was that even computer illiterates walked away with an
appreciation of the Trace series machines. Professor Levine
in a previous article gave Multiflow's address. Since Josh
is going around speaking on this computer I would heartily
recommend taking in his talk if possible, since it is not
only entertaining but also most informative. I have no
affiliation with Multiflow Computer, Inc., other than
setting up a worthwhile seminar by their senior founder. If
you want to email me contact me at the address below since I
am using this account to post this article since our system
administrator hasn't got postnews working on my native
machine.
###############################################################################
# John R. Nickerson #
# 5 King's College Road, Room 214 ## ## ###### ### ### ###### #
# Laser Doppler Anemometry ## ## ## #### #### ## #
# Laboratory Group ## ## ## ## ### ## ###### #
# Department of Mechanical Engineering ## ## ## ## # ## ## #
# University of Toronto ##### ## ## ## ###### #
# Toronto, Ontario, Canada, M5S 1A4 #
# (416) 978-7020 #
# UUCP: {linus, ihnp4, allegra, decvax, floyd}!utcsri!utme!jrn #
# USENET: jrn@me.utoronto.edu #
# BITNET: jrn@ME.UTORONTO #
###############################################################################
# Grad students are like laser beams in #
# a lossless medium, they propagate forever #
# and thus their opinions always diverge. #
###############################################################################