The Intel Pentium CPU experienced a floating-point division flaw, commonly known as the “FDIV Bug,” where certain division operations produced slightly erroneous results. This issue stemmed from a new radix 4 SRT algorithm used in the Floating Point Unit (FPU), which relied on a lookup table. A bug in the script responsible for downloading this lookup table into the hardware Programmable Lookup Array (PLA) caused some entries to be omitted.

Customer impact varied, with Intel initially claiming the flaw would occur once every 27,000 years under normal use, while other reports suggested it could be as frequent as once every 24 days. Intel discovered the flaw in June 1994 but initially chose to keep it private and downplay its significance. Their initial replacement policy was restrictive, requiring customers to prove a need for an unflawed chip, which led to significant public dissatisfaction.

The flaw was independently discovered and publicized by Dr. Thomas R. Nicely in October 1994, leading to widespread public outcry and negative media attention, including a New York Times article in late November. Facing mounting pressure and severe damage to its public image, Intel eventually issued an apology in December 1994 and committed $420 million to a “no questions asked” replacement program for all affected Pentium chips.

Further contributing to the problem, Intel’s FPU testing process had a flaw in its proofs. Engineers incorrectly assumed symmetry in the PLA, leading them to test only half of it. The untested half, which contained the missing entries, was therefore overlooked. This combination of a scripting bug and a flawed testing methodology allowed the defect to go undetected until after production began.