The Intel Chip Flaw That Cost $475 Million

Thomas Nicely, a math professor at Lynchburg College in Virginia, noticed in October 1994 that his Pentium PC was producing slightly wrong answers when computing reciprocals of prime numbers. He isolated the problem to the floating-point division unit and posted his findings to a Usenet group.

The bug was real. Intel's Pentium contained a flaw in a lookup table used by the FDIV instruction — five entries in a table of 1,066 were missing. Under specific inputs, the chip returned a quotient that was off by about 61 parts per million. For most consumer tasks, that error never appeared. For financial models, scientific computing, or anything involving repeated division, it could compound.

Intel acknowledged the bug but adopted a position that would become a case study in crisis communications: they would replace chips only for users who could demonstrate they were doing computationally sensitive work. An engineer at CNN ran the word "Intel" through a simple division sequence live on television. The error appeared. The story became national news.

IBM suspended sales of Pentium-based computers. Analysts began publishing error-rate estimates. Intel's stock dropped. Within weeks, in December 1994, Intel reversed course and agreed to replace every affected Pentium on request, no questions asked. The recall cost the company $475 million — roughly $940 million in 2024 dollars — and the bug's name, FDIV, became shorthand in chip design for the cost of shipping silicon with a mathematical flaw.