A.T. Ho (a@t.ho) on 6/27/07 wrote:
---------------------------
>David Kanter (dkanter@realworldtech.com) on 6/27/07 wrote:
>---------------------------
>>The FDIV bug was uncovered within Intel using either excel or the built in MS calculator.
>>I don't know how the folks outside of Intel found it, but it may be along the same
>>lines. It sounds like most of these issues only impact developers, and are much harder to uncover.
>>
>>DK
>
>If Intel found the bug before anyone else did and didn't disclose the bug,

"The Pentium Chronicles" (nice book) has the full story
(or at least claims to) Basically they found it during P6 validation while cross checking against the older P5.
The P6 people told the P5 people and let them decide.

Their analysis showed that the failure was very rare
and that common numerical algorithms are usually
stable enough to tolerate the errors. So they decided
to not do anything, which ended up being the wrong
thing from the psychological side.

> I'm
>surprised the backlash against them wasn't larger than it was (and there was considerable
>outcry against their efforts to downplay the bug).

x86 CPUs have had worse bugs in fact, it's just
that unlike them FDIV could be easily demonstrated
and shown to the layman using a calculator. Try e.g. to do the same with the F00F bug. That made a big psychological
difference. Even if most people very likely weren't
affected they didn't want to own something they knew
divided wrong because they could understand that.

Anyways, FDIV definitely ended up being a good thing:
Intel always publishes errata regularly since then (and AMD too) which is a big advantage over other architectures where errata are usually secret.