The Pentium Flap

THE great Pentium flap of 1994 may go down as a defining moment, as the computer industry becomes less technology-driven and more consumer-driven.

The Pentium chip is, of course, the microprocessor, the ``brain'' widely used in IBM and IBM-compatible personal computers; it's the fastest, most powerful chip out there.

But over the summer it was discovered that the chip miscalculates ``in floating point divide operations once every 9 billion random number pairs,'' as Intel, its manufacturer, put it.

The design flaw has been corrected, but some 2 million machines are in use with faulty chips. Intel has maintained that this won't matter to most users and, until this week, questioned those seeking replacements to determine whether they really ``needed'' one. It was a major public relations disaster; Intel acknowledged it had seemed ``uncaring.''

Now Intel has announced it will replace flawed chips with no questions asked. Intel's stock rose after the announcement: This is Wall Street's way of granting absolution after repentance has been shown.

Still, Intel could have saved tens, if not hundreds, of millions of dollars by adopting this policy from the start. Volunteering information builds trust: Intel had an opportunity to help change the culture of the computer industry by being more forthcoming with its customers about the problem.

This kind of problem could occur again. The Pentium chip, the size of a fingernail, has 3 million transistors. Chip design is largely automated, but practical difficulties remain in getting new chips fully tested before they are released to the market.

The drive toward further miniaturization will continue. Will manufacturers regard having to maintain generous trade-in policies for flawed chips as part of the cost of doing business on the cutting edge? If so, will high-end users, who really need every last bit of power and miniaturization they can get, accept the resulting higher costs that will inevitably be passed along to them? Will low-end users decide to stick with tried-and-true technologies after all?

We can't ultimately expect our machines to be smarter than we are ourselves.