P54C Erratum 23 -- Slight
Precision Loss for Floating-point Divides on Specific
Operand Pairs
PROBLEM: For certain input datum the
divide, remaindering, tangent and arctangent
floating-point instructions can produce results with
reduced precision.
IMPLICATIONS: Your answers will be wrong.
WORKAROUND: Due to the extreme rarity of
this flaw, a workaround is not necessary for almost all
users.
When this bug was first reported, Intel denied that it
existed. After this bug was proven to exist, Intel denied
that it was a problem. When customers wanted a
replacement chip, Intel demanded that *THEY* (the
customer) prove that they were
affected by this bug. Yet to this very day, Intel refuses
to acknowledge that this is a bug; instead they always
refer to this as a flaw -- whatever the
difference may be.
When I heard about the bug, I didn't think it was any
big deal; and I still don't. But Intel's method of
responded to the bug was a very big deal. Can you imagine
the arrogance of Intel for refusing to honor their
lifetime replacement warranty? This caused the momentum
of public outcry to skyrocket. Intel relented, and
performed an even greater arrogant act: they wanted you,
the customer, to prove that the FDIV bug had affected you
before they would replace your chip. That policy caused
the public outcry to reach the office of Intel's CEO.
Uncle Andy Grove took to the internet in a plea for
public understanding. But nothing seemed to knock the
juggernaut of public outcry off of it's rocker.
The debate about FDIV raged on the internet. I made
the mistake of stating publicly that I didn't think that
the FDIV bug was that big of a problem. In fact, I
defended the existence of the FDIV bug. My stance caused
a huge public outcry of it's own. For some reason, people
thought I was defending the way Intel handled the bug --
which I wasn't. But it seemed clear from this discussion
that many people were under the mistaken impression that
their chip was supposed to be flawless. Sure it's
supposed to be, but in reality that's just a pipe dream.
Any chip with 3 million transistors is going to have
bugs. In this case, it was Intel's handling of the bug
that caused all of the public outcry -- not the bug
itself.
Intel finally relented to the public outcry, and
announced an unconditional replacement policy for any and
all Pentium's affected by this bug. Like I said, I never
thought the FDIV bug was that big of a deal. In fact,
I've never even sent in my Pentium-90 for a replacement.
That should indicate how serious I thought this bug was
to me. Instead, I'll continue to wait for a very long
time before I send in my chip for replacement. My
Pentium-90 has long been end-of-lifed (that means
discontinued). I'll continue to wait until just about
every Pentium-class processor has been discontinued. Then
I'll send in my Pentium-90 for replacement. Maybe Intel
will send me whatever's left in their inventory -- maybe
a Pentium-200 with MMX?
However long it takes, I'll continue to wait and to
dream. But I won't hold my breath for that Pentium-200
with MMX to arrive at my door anytime soon.
|
