what was the deal with the pentium 1 math errors?

[Schadenfroh]

my dad was telling me about it and i rember hearing about it a while back. I was so young at the time i did not live thru it. Would it give you the wrong answer in a spread sheet or if you were designing a bridge in CAD? did the government have to intervine?

 

[JSSheridan]

Do you mean the Pentium rounding error? Link

 

[hifisoftware]

I was taking a programming class at the time. I have spend many hours trying to figure out why a programm that was supposed to show different values for sin() was giving me wrong values... Then I tested it on 486 and it worked perfectly.
You could've easily gotten an wrong answer in CAD. It was guickly dealt in several ways. I think compilers used to have a switch that would produce code that would not run into this issue. Intel also had an offer where they would replace this CPU with one wihtou an error.

 

[AyashiKaibutsu]

Originally, Intel thought it would happen once every 100,000 years, but in reality it happened once every couple of months. Basicly, anyone doing research on them had to redo all of it.

 

[jreg]

I had a 60mhz Pentium that I had to get replaced. It was pretty painless though, I was directed to a local service office (I think the office was for Tandy, heh), where they just took it and gave me another one. I never really ran into any problems with it, but then again I hardly used the faulty one before I replaced it. I can't believe it's already been 10 years since then. heh

 

[Docomo]

As I recall, there was also an even older error with the old Pentium Pros - it was a problem with the divide by 0 command that would or could make the processor lock up. This problem was fixed quickly but not before the French lost an Arianne rocket to the fault - they were using specially designed intel chips that had the error and the onboard computer that controlled attitude couldnt deal with an attitude adjustment request for 0 degrees and the subsequent inflight breakup was spectacular!

 

[n0cmonkey]

Originally posted by: Docomo
As I recall, there was also an even older error with the old Pentium Pros - it was a problem with the divide by 0 command that would or could make the processor lock up. This problem was fixed quickly but not before the French lost an Arianne rocket to the fault - they were using specially designed intel chips that had the error and the onboard computer that controlled attitude couldnt deal with an attitude adjustment request for 0 degrees and the subsequent inflight breakup was spectacular!

F00F?

 

[Accord99]

Originally posted by: Docomo
As I recall, there was also an even older error with the old Pentium Pros - it was a problem with the divide by 0 command that would or could make the processor lock up. This problem was fixed quickly but not before the French lost an Arianne rocket to the fault - they were using specially designed intel chips that had the error and the onboard computer that controlled attitude couldnt deal with an attitude adjustment request for 0 degrees and the subsequent inflight breakup was spectacular!

If you are referring to the failure of the first Arianne 5, it was a software failure relating to the conversion of a 64-bit floating point value to a 16-bit signed integer reused from the Arianne 4 software. Assumptions had been made that were no longer valid with the new rocket and caused an error when value was converted.

 

[KF]

No the government did not get involved. Intel would be nuts to let it go that far, and they aren't nuts. In the end Intel offered to replace the processor for anyone that wanted to. The problem was their first proposal at the big revelatory announcement. They wanted to replace it only for those who could demonstate it was genuinely needed, like in mathematical research. But if I paid the same amount as some mathematician at a university, why can't I get mine replaced too? People forget that the first Pentiums cost over a thousand bucks, and the price did not come down all that fast. They were still getting over $200 bucks for Pentium 1 MMX when PIIs were superceding them.

I doubt if an on-screen CAD display uses the level of precision that would be needed to see the error. I don't think the engineering of bridges use that level of precision either. Accounting for one part in 10,000 is doing very well in real life.

Since it was a floating point error, you probably would never see it in monetary calculations, because they wouldn't use floating point.

Since the error was in the lower bits, and only in some cases, the odds of one calculation being in error were billions to one. However, even back then a computer could do tens or hundreds of millions of calculations in one second. If you crunch for a month or a year, the error odds are not exclusionary. The difficulty is you don't know. Some people did not like the way Intel hammered excessively on how insignificant it was.

It seems to me compilers to this day have an option that works around that bug.
Pretty good account of the bug

"As I recall, there was also an even older error with the old Pentium Pros....This problem was fixed quickly but not before the French lost an Arianne rocket to the fault -"

This was a story intended to illustrate what an error like this could do, but Intel was not involved.

An account of this error

From article: "When talking about the bug Collins gives an interesting analogy about a launch failure of the Ariane 5 rocket, which happened because a floating point to integer conversion overflowed and the overflow was not handled right."