lethin@ai.mit.edu (Richard A. Lethin) (06/06/91)
(This was forwarded to our research group, and I'm surprised to find that it wasn't posted to comp.arch. I don't work for Thinking Machine, but here it is anyway. If you read the New York times article, you've seen this. So, the 64,000-processor question is: What's inside the box?) ----------------------------------------------News Release June 5, 1991 For Immediate Release THINKING MACHINES' NEW CM-200 A 100-USER SUPERCOMPUTER SETS SUPERCOMPUTER SPEED RECORD Cambridge, Mass--June 4, 1991--Thinking Machines Corporation today announced its new 100-user supercomputer, the CM-200, the fastest supercomputer in the world as measured by the industry standard LINPACK benchmark, and the first massively parallel supercomputer to be delivered with a full general purpose software environment. "The CM-200, a production supercomputer system which is available today, has been designed specifically for the network and workstation environment," notes Danny Hillis, chief scientist at Thinking Machines. "Our new CMOST operating environment is UNIX compatible and is the first in the supercomputer industry whose interface is built on the standard X Window System and Motif workstation protocols." Thinking Machines' CM-200, which is compatible with the entire line of Connection Machine supercomputers, clocked 9.03 GFlops (9.3 billion floating point operations per second) double precision and over 17 GFlops single precision on the industry standard LINPACK benchmark. It uses enhanced hardware technology, a faster clock speed, and a new generation of interprocessor communications technology. Peak speed of the CM-200 system is 40 GFlops. Prices for full 64,000 processor CM-200 configurations range from $8 million to $10 million. First deliveries are scheduled for summer. "17 GFlops in LINPACK is extremely impressive," states John Sell, President, Minnesota Supercomputer Center. "But more importantly this is a production system that has the software to support a large user base." CMOST (Connection Machine Operating System: Timesharing) is designed to make programming easier and more efficient. "The CMOST general purpose software environment integrates debugging, performance analysis, and data visualization into a single environment with a familiar industry-standard look and feel. Nobody else does that." adds Ted Tabloski, director of system software at Thinking Machines. CMOST also integrates the acclaimed Connection Machine Fortran compiler, the only supercomputer compiler ever to win the Gordon Bell Prize. "The real measure of the CM-200 is its performance on applications. For example, a Mobil Oil Corporation application which ran at 14 gigaflops on Thinking Machines' CM-2, now runs at 21 gigaflops on the CM-200. We do not know of anyone else in the market today who can sustain 21 gigaflops on a 'real-world' application." said John Mucci, vice president of research at Thinking Machines. The CM-200 consistently delivers 5-10 GFlops of performance to its user community. Among the sites which already report multiple GFlops of sustained performance on their current CM-2's are: Florida State University, Mobil, Los Alamos National Laboratory, and Minnesota Supercomputer Center. Thinking Machines also announced a version of CMOST that runs stand-alone on a Sun workstation. It is a complete Fortran CMOST environment that allows users to write, run, and test Fortran programs right at their workstations. The company will license this package to schools and colleges so that students who do not have access to a Connection Machine supercomputer can independently study parallel programming techniques. Upcoming June announcements from Thinking Machines include the introduction of its aggressively priced entry level CM-200 "Hummer" supercomputer. Thinking Machines also announced today, that it was "on-schedule" for the 1995 delivery of what is expected to be the world's first Teraflop Machine. Thinking Machines Corporation was formed to apply parallel processing techniques to the growing number of data-intensive applications in business and science. Since announcing its Connection Machine system in 1986, the company has grown to become the second largest supercomputer supplier in the United States. Thinking Machines Corporation is headquartered at 245 First Street, Cambridge, Massachusetts 02142. It has subsidiary offices throughout the United States, as well as in Great Britain, France, Italy, Germany, the Netherlands, and Japan. For further information contact: Tim Browne Thinking Machines Corporation 245 First Street Cambridge, MA 02142 617.234.5525 browne@think.com CM-200 Hummer, and Connection Machine are trademarks of Thinking Machines Corporation Unix is a registered trademark of AT&T X Window System is a trademark of the Massachusetts Institute of Technology Motif is a trademark of the Open Software Foundation Sun is a trademark Sun Microsystems, Inc. ------- End Forwarded Message
gillies@m.cs.uiuc.edu (Don Gillies) (06/08/91)
Does each set of 32 processors have 1 FPU? Then each FPU would have to be running at 20 MFLOPS to attain 40Gigaflops ?!! --
wolfe@vw.ece.cmu.edu (Andrew Wolfe) (06/11/91)
In article <1991Jun8.084436.17855@m.cs.uiuc.edu> gillies@m.cs.uiuc.edu (Don Gillies) writes:
Does each set of 32 processors have 1 FPU? Then each FPU would have
to be running at 20 MFLOPS to attain 40Gigaflops ?!!
--
This means a 100ns. Multiply-Accumulate on each node - sound quite reasonable
for todays world.