fpst@hubcap.clemson.edu (Steve Stevenson-Moderator) (03/01/88)
[ I pulled this off the AI bboard. ]
Summary of Spang Robinson Report on Supercomputing and Parallel Processing
Volume 2, No. 1, January 1988
Lead article is on Supercomputer Marketplace
Cray Research
1985 - 100th machine shipped
1987 - 200th machine shipped
Annual market - nine billion today,
twenty billion in 1990
thirty billion in 1995
Academic Supercomputer Application Development
Physical Sciences - 50%
Geosciences - 15%
Biosciences - 8%
Social Science - 2%
Math Sciences - 4%
Engineering - 19%
Multidisciplinary - 4%
Worldwide Distribution of Supercomputers:
U. S. 54%
Japan 19%
Other 8%
UK, France, Germany, Canada less than 5% each
Categorization of where the Supercomptuers are
Research - 24%
Universities 18%
Defense 16%
Aerospace 13%
Petroleum 10%
Environment 7%
Nuclear Energy7%
Service Bureau5%
Automotive 5%
Installed Base (1986):
Supercomputers 228
Vector Augmented 190
Mainframes
MInisupercompute 450
Superminicomputer 140,000
Workstations 110,000
The article also includes estimates for 1991 and beyond for most of the abov
numbers as well as discussion.
______________________________
Article on Thinking Machines Connection Machine.
As of May 1987, they delivered seven systems:
They have raised a total of 31 million in equity capital with a great chunk
untouched.
______________________________
The U. S. House of Representatives Subcommittee on Science, Research
and Technology of the Committee on Science, Space and Technology plans
a brief hearing in the February -early March time frame to review
recommendations for futher support of Supercomputing.
The Office of Science and Technology Policy has a report that
says
1) U. S. government should development long range supercomputer support
program because this is necessary for national security and the economy
2) joint research should be undertaken because of relationship with
software and microelectronic
3) Network technology may become a barrier so further work should
be done linking things up.
4) Long term support should be within available resources:
They estimate $500 million on supercomputing which should be increased
70 percent. The National Network should be upgraded at a cost of
390 million. This includes upgrading existing networks to 1.5 MBit/second
capacity, to be replaced with a 45 Mbit/sec and then to 45 Mbit/Sec.
+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_+_
Shorts:
IBM has made a committment to Steve Chen's company
that falls short of the $100 million needed to develop the product.
Dana Computer has changed its name to Ardent Computer Technology.
Cray Research will be selling a system to Construcciones Aeronauticas, S. A.
Prime Computer Systems is announcing a system developed jointly with
Cydronme. It is a $500,000 =- $600,000 machine using
directed dataflow architecture and vector processors.
A broker, Piper, Jaffray and Hopwood, recommended that investors accumulate
Cray stocks.
Sequent has announced an OEM agreement with Apricot computers of England.
Westinghouse will be selling Scientific Computer System machines in a turnkey
fashion for nuclear fuel customers.
Encore made a profit of $114,000 for the year.
Japan's MITI awarded Tandem its "Good Design Prize for Foreign Products" for its
VLX mainframe and XL8 and V8 disk storage devices.
Encore's President, James R. Pompa has resigned.
Scientific Computers has released a version of System V Release 3 for
its machines which includes UNICOS extensions. They also introduced
a feature to increase the amount of main meory and to implement gather/scatter
and compres instructions.
University of Georgia had an ETA10-P shipped.
BBN Advanced Systems has a low cost starter Butterfly system for $100,000
for universities.
Topologix has introduced add-in boards for Suns containing four transputers
a piece with up to eight boards per workstation.
'Convex has shipped beta site units of C2 which is targetted at 40
MFLOP's.
Indiana University developed a system to interactively automate the conversion
of programs for the Butterfly family of parallel processors. It is public
domain.
Ametek has announced a Series 2010, a hypercube based system, where
messages pass through nodes without requiring processor time.
Each node is a 68020 CPU plus 68881 floating point unit with 1
to eight MBYTES memory. It does not require expansions in powers
of two and can grow to 1024 nodes.
Dow Jones Informations has purchased a Connection Machine from Thinking
Machines. The systems uses the Information retrieval technique
of "relevant factor" in which the user evaluates what is returned from
the first query to generate successive paths.
A second 32,00 processor is on order for March delivery.
One processor has 4.3% of the power of the entire world mainframe
computer power.ephraim@Think.COM (03/02/88)
In article <1036@hubcap.UUCP> fpst@hubcap.clemson.edu (Steve Stevenson-Moderator) writes: >Summary of Spang Robinson Report on Supercomputing and Parallel Processing >Volume 2, No. 1, January 1988 The notes on the Dow Jones project here at Thinking Machines seem to have garbled slightly along the way: >Dow Jones Informations has purchased a Connection Machine from >Thinking Machines. The systems uses the Information retrieval >technique of "relevant factor" in which the user evaluates what is >returned from the first query to generate successive paths. The term more usually employed to describe the system is "relevance feedback." The user evaluates the material returned from an initial query (that's "relevance") and can mark each item as "good" or "bad" (that's "feedback"). >A second 32,000 processor is on order for March delivery. One >processor has 4.3% of the power of the entire world mainframe >computer power. A second 32K (i.e., 32,768) processor *system* is on order. One processor has 4.3% of the power of a plugged nickel. 32,768 processors connected in a system are a force to be reckoned with. Ephraim Vishniac ephraim@think.com Thinking Machines Corporation / 245 First Street / Cambridge, MA 02142-1214 On two occasions I have been asked, "Pray, Mr. Babbage, if you put into the machine wrong figures, will the right answers come out?"
fpst@hubcap.UUCP (Steve Stevenson) (03/04/88)
In article <1036@hubcap.UUCP> you write: >Ametek has announced a Series 2010, a hypercube based system, where <messages pass through nodes without requiring processor time. >Each node is a 68020 CPU plus 68881 floating point unit with 1 <to eight MBYTES memory. It does not require expansions in powers >of two and can grow to 1024 nodes. Ametek series 2010 is not a hypercube based system. Its has a physical connectivity of a 2D mesh with hardware message routing. -- /*------------------------------------------------------------------------*\ | Wen-King Su wen-king@vlsi.caltech.edu Caltech Corp of Cosmic Engineers | \*------------------------------------------------------------------------*/ -- Steve Stevenson fpst@hubcap.clemson.edu (aka D. E. Stevenson), fpst@clemson.csnet Department of Computer Science, comp.hypercube Clemson University, Clemson, SC 29634-1906 (803)656-5880.mabell