Printer Friendly

THINKING MACHINES SHIPS CM-5 VECTOR CHIP SETS NEW STANDARDS FOR PERFORMANCE AND EFFICIENCY

 THINKING MACHINES SHIPS CM-5 VECTOR CHIP
 SETS NEW STANDARDS FOR PERFORMANCE AND EFFICIENCY
 CAMBRIDGE, Mass., Aug. 24 /PRNewswire/ -- Following the public debut of its vector technology at the "Hot Chips" conference at Stanford University, Thinking Machines has formally announced that the new chips are being shipped with installation at customer sites to begin this month. Thinking Machines has already taken delivery of more than 100,000 MFlops of the new chips from vendor Texas Instruments. Destined to have a fundamental impact on general-purpose performance and benchmarks, the chip performs at levels of efficiency that were not previously thought possible.
 "The 1024-node Connection Machine(R) CM-5 will run Linpack in excess of 45 GFlops with these vector units," said Woody Lichtenstein, manager of the company's Fortram Compiler unit. "That's twice the current NEC SX-3/44 record."
 The company emphasized the importance the new chip technology will have for general-purpose computing. "We focused on the most complex applications and algorithms when we designed the CM-5," said Lennart Johnsson, director of computational sciences at Thinking Machines and Gordon McKay, professor of computer science at Harvard University.
 "It is clear from these new performance numbers that our design will meet demand for greatly increased levels of performance and efficiency," Johnsson said. "On a 3D irregular finite element code used in aerodynamics, we have already achieved performance on a 1024 node CM-5 that is seven times faster than the largest CM-2. On a dynamic mesh climate model application, it is more than 23 times faster." The company's CM-2 is already the reigning champion in the IEEE contest as the fastest supercomputer in the world.
 Because the new chips have an extremely high level of efficiency on mathematical applications, users will make dramatic gains in functionality. Each parallel node of a CM-5 can have four vector units, each with a peak speed of 32 MFlops. On critical mathematical operations such as matrix-vector multiply, the system, as a whole, runs at 61 percent of peak. For the portions of the operation that take place within the node itself, the system achieves more than 95 percent of the peak, or 121 MFlops.
 "We had the advantage of years of experience with parallel programming and compilers," said Jon Wade, architect and project leader of the company's vector units. "The high efficiencies that we are achieving come directly from this understanding of the systems context. And they prove that the CM-5 architecture will indeed sustain a TFlops of performance."
 Thinking Machines Corporation is headquartered in Cambridge, with offices throughout the United States, Europe, Australia and Japan. The company is privately held.
 NOTE: Connection Machine is a registered trademark of Thinking Machines Corporation.
 -0- 8/24/92
 /CONTACT: Glenda Cudaback of Thinking Machines, 617-234-5501/ CO: Thinking Machines Corporation ST: Massachusetts IN: CPR SU: PDT


TM -- NE009 -- 2559 08/24/92 11:41 EDT
COPYRIGHT 1992 PR Newswire Association LLC
No portion of this article can be reproduced without the express written permission from the copyright holder.
Copyright 1992 Gale, Cengage Learning. All rights reserved.

Article Details
Printer friendly Cite/link Email Feedback
Publication:PR Newswire
Date:Aug 24, 1992
Words:472
Previous Article:PETER HARRIS SUCCEEDS DONALD WOOD AS PRESIDENT ARCO CHEMICAL ASIA PACIFIC OPERATIONS
Next Article:'IT'S TIME TO EXPECT MORE FROM A CAR' IS NEW NISSAN DIVISION SLOGAN
Topics:


Related Articles
CREE RESEARCH UNVEILS NEW SUPER BRIGHT BLUE LED; COMPANY IS WORLD'S FIRST SUPPLIER OF BRIGHT BLUE CHIPS
Crusoe Finally Reaches Civilization. Is It Too Late? Does Civilization Care?
Analog Devices' DashDSP Powers Toshiba-Carrier's New Air Conditioner Units.
BEA Demonstrates Market's Fastest Benchmark Results on BEA JRockit Using Dual-Core Intel(R) Xeon(R) Processors.
New Cree GaN HEMT Optimized for North American WiMAX Applications.
Cree's New GaN Transistors Boost WiMAX Power Amplification Efficiency by up to 50 Percent.
Apple Updates Xserve with Twice the Performance.
AMD plans 16-core server chip -- but not until 2011.
AMD plans 16-core server chip -- but not until 2011.

Terms of use | Copyright © 2017 Farlex, Inc. | Feedback | For webmasters