CASE: PENTIUM FLOATING POINT DIVIDE

In October, 1994, Professor Thomas Nicely, a professor of mathematics at Lynchburg College in Virginia, notified Intel Corporation about an apparent flaw in the Intel P5 Pentium chip. The flaw had to do with floating point division. According to Professor Nicely, the division of one floating point number by another floating point number sometimes produced invalid results. These invalid results only occurred under certain circumstances and were unlikely to be noticed by most users.

While this problem might not seem significant, it proved to be a serious issue, affecting most of the numerical calculations performed by PCs at that time. Even though the erroneous values were significant only for those calculations requiring high precision and were quite rare (approximately once in every nine billion floating point divide operations), the total number of such operations that were done every day by PCs was enormous, so that the number of faulty values was also very large.

According to Nicely, people at Intel indicated they had been aware of the problem for several months, but had not announced it to the general public. The following month, a report on CNN turned the bug into real news. Intel argued that the bug was not very important and affected very few users. The company offered to replace chips only for those users who could demonstrate a need for high precision. Public outrage ensued, with critics arguing that Intel had an obligation to deliver a product that behaved as it was supposed to. They also argued that users should not be required to predict their future needs for precision, and instead should be able to count on their computers to provide the precision defined by current computational standards as defined be standards agencies. Soon, probably because of the public outcry, Intel reversed its position and agreed to replace all of the flawed chips.

The cost to Intel for the floating point divide bug has been estimated at approximately one-half billion dollars. The loss of trust by the public is difficult to measure in dollars, but there can be no doubt that Intel’s image was tarnished by the bug and by the company’s initial reluctance to disclose the problem and replace the faulty chips.