An episode in machine precision history...


Well-Known Member
Licensed User
I found this posting interesting :

because of this bit of INTEL history....
[[found in "the intel trinity" :Chapter 52 : "The bug of all bugs : author : Micheal S Malone ISBN 978-0-06-222676-1]]

(I know several people who on getting new generation chip tech, brow-beat them with what seem to be unvarifiable testing related to this story.)

will Try to keep it short.....

This incident nearly took INTEL out due to the sloppy Pubic Relations response.

It goes like this.

The story of a college professor, from Virginia, called Thomas Nicely is told, as he, evidently, had taken advantage of Moore's law (486 to Pentium tech) to get some real number crunching done :

"On June 13, 1994, in the middle of running several billion calculations on his new Pentium-based computer, Nicely was stunned to find an error in his results. Where 4,195,835 divided by 3,145,727 should have given the answer of 1.333820449, the computer instead read 1.333739068. How could that be? he asked himself. Computers don't make mistakes like that." (pg 440)

After four months of reworking his calculations, Nicely concluded that the Pentium was flawed. INTEL told him he was the only person out of 2 million Pentium users to have detected this.

The company made a note, but, apparently treated this as in the normal course of things; minor bugs exist in new chips, and, waited for more evidence, leaving an if necessary fix for the next generation of the product.

Except : Nicely couldn't believe no-one had encountered the flaw and asked his connections in the science world whether anyone had encountered it. They couldn't believe it and posted to Compuserve's members to see if anyone could add anything.

At this point an early viral effect must have happened (remember the Internet was still a young and immature wire-works at the time). No less than the Wall Street Journal was among the media house to pick up the story and INTEL became a Big Giant stonewalling Little David facing a big ruckus in the media.

By November 2, several lawsuits had been brought against INTEL.

INTEL had not totally ignored the report, when it received a few more complaints, it launched a low-level investigation, but still with the thinking that it was in-business-as-usual mode. The world had changed since the 486-days four or five years earlier.

What happened next, [and this is weird for me, how did they come up with this lot] as Malone goes on,
"...[INTEL] investigated the problem and found that it was real, but so rare as to not merit any immediate attention. INTEL's investigators determined that it was a rounding error that only occurred once in every 9 billion division operations. Nicely had run into it because he was crunching such a vast number of divisions. The average spreadsheet user, INTEL determined, might run into this every 27,000 years. That made it so rare as to be immaterial......". (pg 442)​

The rest of the story is about a PR disaster, and consumer anger, and loss of trust in a big corporate (did you read this FACEBOOK?) : get the book if you can, its a really interesting business history (and convince yourself that computers are not about physics but chemistry; in case you weren't on that page).

Oh, and, if you want to calculate Pi accurately : do it with infinite precision using long division and pencil and paper. :rolleyes: