Intel To Users: 'Humbug!'

Welcome Guest. | Log In| Register | Membership Benefits

Intel To Users: 'Humbug!'

Though the chip maker says flaws in its Pentium chips are minor, user complaints are mounting

By Stephen Manes
Issue column appeared: Dec. 19, 1994

Intel Corp.'s response to the flap over its Pentium processor bug is turning into a textbook example of how not to handle a delicate situation. Angry customers have posted hundreds of complaints on CompuServe and the Usenet. These users, who paid top dollar for the most expensive microprocessor in its class, have learned that the chip doesn't handle floating-point division consistently. When the users request help, the Intel "sysbot" repeats, "If you have questions about the subtle floating-point unit flaw of the Intel Pentium processor, cal l 1-800-628-8686."

Not A Subtle Flaw
Despite Intel's assurance that "an average spreadsheet user could encounter this subtle flaw once in every 27,000 years of use," the flaw turns out to be less subtle than first reported. Want to reproduce the bug on your own system?

Try the formula (x/y)*y-x, with x = 4195835 and y = 2.9999991. It should, of course, work out to zero. But on a buggy Pentium--which now means most of them--you get negative 256. Similar miscalculations with other "magic numbers" quickly convince skeptics that something is seriously wrong.

The most famous floating-point bug appeared in 1982, when a consultant discovered that the Basic programming language in the original IBM PC didn't perform certain decimal calculations quite right. That bug, courtesy of Microsoft, was burned into a ROM in every machine. Software workarounds provided an acceptable fix.

But that happened at the beginning of the PC revolution, when perhaps 200,000 machines were affected and lit tle software had been written for them. With Pentium, we're talking about countless programs and more than 4 million chips. Software vendors seem unlikely to want to waste resources on writing, recompiling, and distributing workarounds that their customers won't want to pay for.

The angriest users are those whose jobs rely on millions of iterations of calculations, where small errors can easily snowball into big ones. It's not just the bug that rankles; it's the perception that any work done on a Pentium might be inaccurate. The perception is not unfounded: On CompuServe, you can practically hear people slapping their foreheads as they discover why their old 486-based calculations of circuits and finances come out differently on their blazing new Pentiums.

Average users are indignant, too. They realize their computers are now damaged goods with declining resale value. They know Pentium doesn't cause obvious system crashes; it generates tiny but significant errors that may go undetected until too late.

While Intel president Andy Grove now says he's "truly sorry," the company pooh-poohed user concerns at first, claiming its chip problem would affect "users of the Pentium processor who are engaged in heavy-duty scientific/floating-point calculations," as Grove's Internet posting put it. Users, upon reaching Intel's 800 number, apparently go through a lengthy interview process to see if Intel deems them worthy of receiving a corrected chip. If you can't convince Intel that you may encounter the bug in daily life, you just don't make the cut.

The Complexity Iceberg
Intel may have to rethink this policy soon. The term "class action lawsuit" keeps popping up around the Net. Customers vow to sit on their hands until vendors can guarantee them the bug-free chips expected to trickle into the market.

But no matter how Intel responds, this problem will remain merely the tip of an iceberg called complexity. Grove's Internet posting makes this clear: "Because this chip is three times as co mplex as the 486, and because it includes a number of improved floating-point algorithms, we geared up to do an array of tests, validation, and verification that far exceeded anything we had ever done." Obviously, it wasn't enough.

The still-greater complexity of Intel's forthcoming P6 chip means Intel will face an even tougher testing job--precisely at the same time it faces increased competition and hopes to say, as Grove did about the Pentium, "We ramped the processor faster than any other in our history." It's this sort of burgeoning, almost untestable, complexity in the face of pressure to deliver that haunts nearly every software developer.

At least Pentium owners have one reason to smile: Their chip has prompted a hilarious list of "Top Ten Reasons to Buy a Pentium Machine" on the comp.sys.intel forum.

Reason No. 10.0000001: "Your current computer is too accurate."

To see our most recent Pentium flaw news story, click here