🚀 go-pugleaf

at455@FreeNet.Carleton.CA (Colin Pascal) writes:

>In a previous posting, Ian Staples (ianst@qdpii.ind.dpi.qld.gov.au) writes:
>> The latest edition of _PC Week_ in Oz had a test for the crook chips on p.1:
>> 	A1=4195835
>> 	A2=3145727
>> 	A3=A1-(A1/A2)*A2

[Note that the algebraic answer is A3=0, bad Pentiums say A3=256]

>Okay, I have a pentium, and I just have one question: How is this going to
>affect me?  I have not had a single error so far.  When does this bug come
>into play?

As someone else said:

>On the other hand, If I'd done five months research using a brand
>new PC with a Pentium and found out that my results were potentially
>invalid (even if it was a 1 in a zillion chance, you still have to
>retest!), I'd be a mite pissed.

Yeah, the whole point is that it *completely* destroys public faith in
computers [not that a lot of this isn't misplaced anyway :) ] -- people
will be arguing over bank accounts, phone bills, business invoices, etc.,
etc.  I mean, we all hope to win at Lotto one day, which has about the same
odds, so why shouldn't we be equally worried about losing something due to
a computer fault, even if rare?

I would imagine all banks, other businesses, accountants, Universities and
all other research establishments, architects, civil/mining/mechanical/etc.
engineers, CAD/CAM, process control applications, etc. etc. etc. could not
possibly tolerate *knowing* that some of their calculations *will* be
wrong in the long run.

As the test example showed, arguments about decimal places are irrelevant
when you are considering differences between numbers of large magnitude
and it's the *differences* that are important (e.g. error mean squares
in stats, profit/loss in business, resultant vectors in engineering ... ).

As a matter of interest, does anyone know how many number combinations are
affected by the bug?  I would imagine there is some readily defined series
of number pairs which would demonstrate the error.

Also, when did Intel start shipping fixed chips?  Two of my colleagues here
in Oz bought PCs with bad Pentiums quite recently (the Biometrician is *very*
pissed off indeed, as one would expect in view of his trade in numbers).

I wonder if Intel will survive this?  Surely it can't be long before people
start suing, especially as it seems Intel were aware of the problem before
the news broke publically -- imagine if they had been selling a pharmaceutical
with undisclosed, albeit rare, potentially serious side effects!

Somehow I'm glad I've procrastinated on upgrading my 386SX-16 :-)
[But of course I don't run Windows unless I *really* have to :) ]

Ian S.

-- 

Ian Staples                        E-mail : ianst@dpi.qld.gov.au
c/- P.O. Box 1054 MAREEBA          Phone  : +61 (0)70 921 555 Home 924 847
Queensland Australia 4880            Fax  : +61 (0)70 923 593   "   "   "