---
title: "Pentium FDIV Bug: The Chip That Got Division Wrong — Cloud Changelog"
description: "In 1994 a professor found that Intel's Pentium got some division wrong. The five missing table entries, the email that spread it and the $475M charge."
url: "https://cloudchangelog.com/articles/pentium-fdiv-bug"
updated: "2026-11-19"
---

# Pentium FDIV Bug: The Chip That Got Division Wrong

A math professor found that Intel's Pentium chip got some division wrong, and Intel's first answer made things worse.

November 19, 2026

On October 4, 1994, a mathematics professor in Virginia ran the same sum on two computers and got two different answers. Thomas Nicely taught at Lynchburg College, and his research meant counting prime numbers, a job so big that he spread it across a group of computers. In March he had added a Pentium, a chip from Intel, to the group. When he compared the Pentium's total with the total from an older machine, the two were far apart.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1)

## Blaming his own code

It was not the first time the numbers had looked wrong. In June he had noticed that one of his counts did not match a published figure. He spent the summer searching through about 3,000 lines of his own code for the mistake. He did find a real flaw, in the software that translates a programmer's code for the computer, worked around it, and rewrote parts of his program. In September he put the revised program to work. A few weeks later it produced another result that did not add up.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1)

This time he hunted more carefully. By trial and error he tracked the mismatch to a pair of huge numbers. Dividing one by the first of them, 824,633,702,441, gave a wrong answer on the Pentium. A second Pentium computer made the same mistake. So did a third, a Packard-Bell on display at an office supply store. The older 486 machines he tried did not.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1)

In October he called Intel's technical support. A support contact there later told him that Intel had seen the same error on one of its own Pentiums, but had no explanation, and that nobody had reported such a bug before.[\[2\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-2) He also mailed a test program on a floppy disk to the company that built his computer. No reply ever came.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1)

## Five empty cells

The cause turned out to be a table. Intel had been stung by rival chips that were quicker at this kind of arithmetic, so its engineers gave the Pentium a faster method of division. The method builds each answer in bigger steps and consults a built-in table that was meant to hold 1,066 entries.[\[3\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-3)

An engineer had prepared the table on a computer and written a short program to copy it into the chip's design. A mistake in that program meant five of the entries were never copied, and nobody checked the finished design to see that the copy was right. Where those five cells should have held a plus two, the chip found a zero.[\[3\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-3)

Most divisions never touched those cells. When one did, the answer came out slightly off, usually around the ninth or tenth digit, and a random division hit the problem about once in nine billion tries.[\[3\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-3)

The worst known case was found by another researcher, Tim Coe. Divide 4,195,835 by 3,145,727, multiply the result by 3,145,727, and take that away from 4,195,835. The answer should be zero. On a flawed Pentium it was 256. Even the Windows calculator could run the sum.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1)

## An email and a chain of strangers

On October 30, with no real answer from Intel or the computer's maker, Nicely wrote to people he thought might have access to Pentiums.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1) His message gave a test anyone could try: multiply 824,633,702,441 by one divided by 824,633,702,441, which should give one, give or take a tiny rounding error. The Windows calculator in scientific mode would do. Every Pentium he had tested failed, from five different makers, and no 486 or older machine did. He said anyone could use his findings freely as long as they gave his name and his employer, and he asked for test results from other chips.[\[2\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-2)

Of the magazine writers on his list, only one answered, to say he would pass it to his magazine's testing lab.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1) That was Tom Halfhill of Byte, who later wrote an account of what went wrong.[\[3\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-3)

A software author on the list, Andrew Schulman, forwarded the message two days later to Richard Smith, who ran a software company in Massachusetts. Smith's company quickly confirmed the error with the Windows calculator. He emailed Intel and several makers of programming tools, then posted the message on a CompuServe forum and asked people to try it. About ten confirmations came back within a day.[\[4\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-4)

A reporter for Electronic Engineering Times saw the post and began asking around. In November, Terje Mathisen, who worked at a company in Norway, confirmed the error and posted the result to a public discussion group, which started extensive discussion online. On November 7 the magazine put the story on its front page, reporting that Intel had confirmed it had changed the part of the Pentium that does this arithmetic.[\[4\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-4)

Intel, it turned out, already knew. Five days before that front-page story, a program manager there told Nicely that his analysis was essentially correct, that Intel had noticed the problem in its own testing, and that a fixed version of the chip was already out in sample quantities. Nicely later learned this had apparently happened in May or June. The manager offered to send two replacement chips, and they arrived the next day.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1)

Later in November, a CNN reporter became the first mainstream journalist to tell the story. Other national and international news outlets followed.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1)

## Once in 27,000 years

Intel's answer was that almost nobody would ever notice. In November it published a technical paper that called the flaw of no concern to the vast majority of users. Only a few users in scientific, engineering and financial engineering work, who need unusual precision and run millions of divisions a day, might need a fixed chip or a software workaround, the paper said.[\[5\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-5)

Intel's estimate was that a typical spreadsheet user might meet the flaw once in 27,000 years. That figure assumed about a thousand divisions a day and numbers that landed on the bad cases only by chance. Intel checked it against 510 of its own spreadsheets.[\[3\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-3)

IBM disagreed and released a study of its own on December 12.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1) The study said a spreadsheet user could hit the flaw once every 24 days, because it assumed about four million divisions a day and bad cases that turned up more often than chance. IBM also pulled its Pentium computers out of stores.[\[3\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-3)

Halfhill thought the dispute would probably never be settled, since both estimates rested on assumptions about how people use spreadsheets.[\[3\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-3) Nicely thought Intel's side was closer to the truth. He called IBM's study a serious exaggeration. He also wrote that a competitor of Intel had chosen to publicize an estimate he believed was too high.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1)

## No questions asked

Until then, Intel had offered a replacement only to owners who could show they needed one. Many customers were furious, and Intel became the butt of jokes. On December 20, Intel gave in. Anyone who asked would get a free replacement, with no questions. Andrew Grove, Intel's chief executive, told industry analysts on a call: "To some people, this seemed arrogant and uncaring. We apologize for that."[\[6\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-6)

Grove still did not think the flaw mattered much. He called it "an extremely minor technical problem" that had taken on a life of its own, and Intel still put the chance of hitting it at once in 27,000 years. It was changing its policy, he said, so that nobody would doubt Intel stood behind the product. Intel's shares rose that morning, to $60.50, as investors hoped the trouble was over. Owners could have a corrected chip sent to them or take their computer to a dealer, and the chip would be widely available within a few months.[\[6\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-6)

On January 17, 1995, Intel said it was taking a $475 million charge against its earnings to cover replacing the flawed chips.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1)

A newspaper at the end of that month noted that the shares had kept climbing anyway, and that Intel was expected to ship about 27 million Pentiums that year, five times as many as in 1994.[\[7\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-7)

Nicely thought the recall was good public relations but a poor use of money. He wrote that fewer than one user in ten had any real reason to worry about the flaw, and that fixing it was not worth more than 100 million dollars.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1)

## The software fix

For programmers who could not wait for new chips, there was a software fix, though it only helped in programs that were rewritten and shipped again. Intel's recommended workaround was to replace each division in a program with a small check. If the numbers were among the troublesome ones, the program adjusted both of them first, which avoided the bad cells and left the answer the same. The check slowed a program by perhaps a fifth, depending on the program. Turning off the part of the chip that does this arithmetic also worked, but it could slow programs badly, and some would not run without it.[\[1\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-1)

Halfhill's view in Byte was that the Pentium did not have a defect like a faulty appliance. It had a bug of the kind any programmer would know, written into silicon instead of shipped on a floppy disk. What set it apart was that it did not crash the computer. It handed back wrong answers so slight that a user might never notice.[\[3\]](https://cloudchangelog.com/articles/pentium-fdiv-bug#source-3)

## Sources

1. Thomas R. Nicely, Lynchburg College (Internet Archive copy) — “[Pentium FDIV flaw FAQ](https://web.archive.org/web/20120125142121/http://www.trnicely.net/pentbug/pentbug.html)” (August 2011). [↩](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-1)[↩2](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-2)[↩3](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-3)[↩4](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-4)[↩5](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-5)[↩6](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-6)[↩7](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-7)[↩8](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-8)[↩9](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-9)[↩10](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-10)[↩11](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-11)[↩12](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-12)[↩13](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-13)[↩14](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-1-14)
2. Thomas R. Nicely, Lynchburg College (Internet Archive copy) — “[Original Pentium FDIV flaw e-mail](https://web.archive.org/web/20120125130740/http://www.trnicely.net/pentbug/bugmail1.html)” (October 1994). [↩](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-2-1)[↩2](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-2-2)
3. BYTE, Tom R. Halfhill (Internet Archive copy) — “[The Truth Behind the Pentium Bug](https://web.archive.org/web/20060209005434/http://www.byte.com/art/9503/sec13/art1.htm)” (March 1995). [↩](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-3-1)[↩2](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-3-2)[↩3](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-3-3)[↩4](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-3-4)[↩5](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-3-5)[↩6](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-3-6)[↩7](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-3-7)[↩8](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-3-8)
4. Richard M. Smith, Phar Lap Software (Internet Archive copy) — “[How Dr. Nicely's Pentium FDIV bug report reached the Internet](https://web.archive.org/web/20120125115647/http://www.trnicely.net/pentbug/rsmith.html)” (December 1994). [↩](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-4-1)[↩2](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-4-2)
5. Intel (Internet Archive copy) — “[Statistical Analysis of Floating Point Flaw: Intel White Paper - November](https://web.archive.org/web/20080318174409/http://www.intel.com/support/processors/pentium/sb/CS-013005.htm)” (November 1994). [↩](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-5-1)
6. The Associated Press via Deseret News — “[Intel surrenders over flawed Pentium chip](https://www.deseret.com/1994/12/20/19149013/intel-surrenders-over-flawed-pentium-chip/)” (December 1994). [↩](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-6-1)[↩2](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-6-2)
7. Knight-Ridder via The Spokesman-Review — “[Intel Stock Climbing Despite Flap Over Pentium Chips](https://www.spokesman.com/stories/1995/jan/29/intel-stock-climbing-despite-flap-over-pentium/)” (January 1995). [↩](https://cloudchangelog.com/articles/pentium-fdiv-bug#cite-7-1)

## More articles

- [Mars Climate Orbiter: How a Unit Mix-Up Lost a Spacecraft](https://cloudchangelog.com/articles/mars-climate-orbiter-unit-mix-up)

  A NASA spacecraft flew for nine months, then was lost within minutes of reaching Mars, because one file used pounds where everyone expected metric units.
- [A Short History of the Changelog](https://cloudchangelog.com/articles/history-of-the-changelog)

  How a plain list of software changes went from a programmer's private diary to app store jokes and pages that investors follow.
- [The Worst Software Rollouts of All Time](https://cloudchangelog.com/articles/worst-software-rollouts)

  Six launches and updates that went wrong within hours, told in the order they happened, and what each one cost.

Get a free daily, weekly or monthly summary by email.

## Contact

- [YouTube](https://www.youtube.com/@cloudchangelog)

```json
{
  "@context": "https://schema.org",
  "@type": "Article",
  "author": {
    "@type": "Organization",
    "name": "Cloud Changelog"
  },
  "citation": [
    {
      "@type": "CreativeWork",
      "name": "Pentium FDIV flaw FAQ",
      "publisher": {
        "@type": "Organization",
        "name": "Thomas R. Nicely, Lynchburg College (Internet Archive copy)"
      },
      "url": "https://web.archive.org/web/20120125142121/http://www.trnicely.net/pentbug/pentbug.html"
    },
    {
      "@type": "CreativeWork",
      "name": "Original Pentium FDIV flaw e-mail",
      "publisher": {
        "@type": "Organization",
        "name": "Thomas R. Nicely, Lynchburg College (Internet Archive copy)"
      },
      "url": "https://web.archive.org/web/20120125130740/http://www.trnicely.net/pentbug/bugmail1.html"
    },
    {
      "@type": "CreativeWork",
      "name": "The Truth Behind the Pentium Bug",
      "publisher": {
        "@type": "Organization",
        "name": "BYTE, Tom R. Halfhill (Internet Archive copy)"
      },
      "url": "https://web.archive.org/web/20060209005434/http://www.byte.com/art/9503/sec13/art1.htm"
    },
    {
      "@type": "CreativeWork",
      "name": "How Dr. Nicely's Pentium FDIV bug report reached the Internet",
      "publisher": {
        "@type": "Organization",
        "name": "Richard M. Smith, Phar Lap Software (Internet Archive copy)"
      },
      "url": "https://web.archive.org/web/20120125115647/http://www.trnicely.net/pentbug/rsmith.html"
    },
    {
      "@type": "CreativeWork",
      "name": "Statistical Analysis of Floating Point Flaw: Intel White Paper - November",
      "publisher": {
        "@type": "Organization",
        "name": "Intel (Internet Archive copy)"
      },
      "url": "https://web.archive.org/web/20080318174409/http://www.intel.com/support/processors/pentium/sb/CS-013005.htm"
    },
    {
      "@type": "CreativeWork",
      "name": "Intel surrenders over flawed Pentium chip",
      "publisher": {
        "@type": "Organization",
        "name": "The Associated Press via Deseret News"
      },
      "url": "https://www.deseret.com/1994/12/20/19149013/intel-surrenders-over-flawed-pentium-chip/"
    },
    {
      "@type": "CreativeWork",
      "name": "Intel Stock Climbing Despite Flap Over Pentium Chips",
      "publisher": {
        "@type": "Organization",
        "name": "Knight-Ridder via The Spokesman-Review"
      },
      "url": "https://www.spokesman.com/stories/1995/jan/29/intel-stock-climbing-despite-flap-over-pentium/"
    }
  ],
  "dateModified": "2026-11-19T19:40:00.000Z",
  "datePublished": "2026-11-19T19:40:00.000Z",
  "description": "In 1994 a professor found that Intel's Pentium got some division wrong. The five missing table entries, the email that spread it and the $475M charge.",
  "headline": "Pentium FDIV Bug: The Chip That Got Division Wrong",
  "image": "https://cloudchangelog.com/_files/3LKjxEzy7iiSV3FYehOlOemjcXv/hH-HFXZ6EJj3zM3sYTUPSA/original.png",
  "mainEntityOfPage": "https://cloudchangelog.com/articles/pentium-fdiv-bug",
  "publisher": {
    "@type": "Organization",
    "name": "Cloud Changelog"
  }
}
```
