In 1994, Intel released a microprocessor that had a major flaw that they dismissed as simple and not likely to replicate very often while in use. The Pentium microprocessor, used in both the 486DX and Pentium CPUs both included a floating point unit, also known as a math coprocessor, which was used to help spreadsheet users. Floating point units used non-integers in their arithmetic methods instead of integers, much like the previous CPUs. FPUs made the arithmetic must faster and allowed for more complex calculations. However, the problem was that there was an error in the FPUs calculations.
The error, which Intel said would show up so little that they were quoted saying “once in 27,000 years” actually showed up ...view middle of the document...
33374 to 6 sig figs instead of 1.33382 to 6 sig figs, which is a relative error of 0.006%. Seems small, yes, but for a quality product to supposedly make things easier, an error is not what you want to see.
The news got a hold of this when Thomas Nicely, a math professor at Lynchburg College in Virginia, was computing sums of reciprocals of a large collection of prime numbers. He got incorrect results on the Pentium computer but correct results on an old 486 CPU computer. He inquired to Intel, got no real response, and posted a general notice on the Internet asking for others to confirm his findings. The media snatched up the information and story quickly, and Intel had to respond. Within a month, IBM halted shipment on these Pentium-based computers, Intel set up a support group to replace the microprocessor for any owner who asked for one.
Both a good move and a bad move on Intel’s part, because they should have caught the problem early and admitted to it when it first appeared. Especially when a group on the internet presented replicating problems that anyone could show. You can’t say it doesn’t happen when someone has irrefutable proof to lay in front of you. That’s like saying the Holocaust never happened when there are pictures blatantly showing those atrocities, though of course they are not nearly on the same scale of horrid. All things aside, Intel did the right thing by replacing the microprocessors to any who wanted them replaced. It might have cost them $500 million, but at least they fixed their mistake in the end.
Janeba, M. (1994). The Pentium Problem. Retrieved 01 28, 2014, from Willamette's Department of Mathematics: http://www.willamette.edu/~mjaneba/pentprob.html