Roanoke Times
Copyright (c) 1995, Landmark Communications, Inc.
DATE: SATURDAY, December 24, 1994 TAG: 9412270081
SECTION: NATIONAL/INTERNATIONAL PAGE: A-1 EDITION: METRO
SOURCE: Associated Press
DATELINE: NEW YORK LENGTH: Medium
A physics professor in Louisiana may have to scrap a semester's worth of work. An aggravated grad student in Massachusetts is reappraising communications research. A health consultant in Georgia frets that the computer software he created for clients can't calculate exactly right.
And at Los Alamos National Laboratory, birthplace of the atomic bomb and center of U.S. nuclear weapons research, bosses are concerned enough to have set up a ``Pentium hot line.''
They've all been bitten by the Pentium bug.
The flaw in the coin-size silicon brain that powers millions of personal computers has garnered enormous national attention because so many PCs have invaded American family rooms. But the practical effects of the Pentium defect are felt most in university labs, businesses and government agencies.
In rare instances, the Pentium boo-boo can botch an obscure decimal-point division calculation. Intel Corp. of Santa Clara, Calif., which makes the Pentium and has flaunted its computing might, has said ordinary computer users will never notice.
That's not good enough for researchers, angered and anxious over completed work they had assumed was untainted. Many now feel they must recheck research done on Pentium-powered PCs, including such well-known brands as Compaq, Dell, Gateway and Packard-Bell.
This week, Intel abandoned its rigid prove-you-really-need-a-new-chip attitude and announced free Pentium replacements for all customers, more than 2 million worldwide. At the top of its priority list: aerospace, medical, scientific and financial institutions, including schools and government agencies.
``When we first heard there was a possible flaw, we immediately became concerned because the kind of calculation it affects, floating point division, is something we do quite often,'' said Juan Rodriguez, physics professor at Centenary College in Shreveport, La.
His class had performed simulations of organic molecule behavior all semester long on a Pentium-powered PC. When the flaw was disclosed last month, the class retried one experiment on another computer and produced different results.
``I am angry,'' Rodriguez said. ``I've lost three months worth of work and my students have as well. My students were looking forward to this upcoming paper where all their efforts would be documented. Now that's postponed.''
The famed laboratory at Los Alamos is replacing 1,000 Pentium chips in its computers, but the flaw poses no risk of misfire or meltdown to the nation's nuclear arsenal. Critical computing is done on more sophisticated systems, the lab said Friday, with the Pentium-powered units relegated chiefly to accountants and secretaries.
``Our major simulations are done on mainframes, not on PCs,'' said spokesman John Gustafson. ``That includes simulations in the weapons field and global climate models.''
The flaw, a missing instruction in the chip, doesn't affect things consumers find most useful and fun about computers, such as word processing and games. It can, however, produce a wrong answer in division problems that involve 1,700 combinations of numbers, such as 5,505,001 divided by 294,911.
For each of the 1,700 combinations that could go wrong, 9 billion others can be divided accurately.
Murat Salih, who is pursuing a doctorate in electrical engineering at the University of Massachusetts, said his Pentium PC cranks through complex equations 17 hours a day. Just one component of the simulations he performs in researching the behavior of audio signals can require 200,000 calculations.
``The trouble is when you are doing numerical simulations, it's very hard to find the error,'' Salih said. ``It's not like you check every calculation.''
A freelance designer of analytical software for a health care management firm in Atlanta said he hadn't found any problems but asked for a new chip anyway because of the challenge of checking for mistakes.
``There may be such a thing as a hardware bug and you could see those kind of problems,'' said developer R. Bolin, who asked to be identified only by first initial. ``But when you're dealing with numbers, you just expect it to be accurate.''
The first public disclosure of the flaw came in late October when Thomas Nicely, a math professor at Lynchburg College in Virginia, sent e-mail to some colleagues about a deviation in computations he had found.
His note eventually wound up on the Internet, the global computer network that has become a sort of perpetual high-tech kaffeeklatsch.
``The most important thing we should have done is we should have been more in touch with the community of people on the Internet and the online services,'' Andrew Grove, Intel's chief executive officer, said this week. ``I think we would have gauged the significance of the problem much more if we had.''
by CNB