#Pentium Probs
48 messages in this thread
"Considering the problems with the Pentium"
The Pentium problems consist of five missing entries in a floating
point lookup table. Unless the application is doing a huge amount of
floating point calculations, the "problem" is essentially a non-issue
in practical terms for the vast majority of users.
Intel, of course, did a horrible job of dealing with the flaw and
it's understandable that every user wants a "perfect" Pentium (which
is why I've already placed a call to Intel's replacement line).
But there's no reason to overstate the flaw as if it is some
crippling defect. It isn't.
"Unless the application is doing a huge amount of floating point
calculations,"
Like ray tracing for instance? 🙂
or just a tad more important, like stress/structural analysis!!!
Matt Straney, 3 hours late, without your luggage
from Grand Ledge — on AutoPilot ——–>
"Like ray tracing for instance?"
That was my initial thought. But you'd have to hit numbers that attempted to
access those five missing floating point data table entries – and even then the
numbers would have to be in certain combinations before you would even see
their impact.
No doubt about it: Intel owes everyone with a Pentium CPU a new,
corrected unit. But it's hardly an "end of the world" situation.
Intel is now providing replacements to anyone who requests one –
after their first blundering announcements.
I just don't see it as a big deal, although it does provide the
basis for a lot of good jokes. <G>
Jerry
Here's an easy calculation to add to the list:
4.999999 / 14.999999 * 3.000001
Which is one point something…. The Pentium thinks it's less than one.
What *I'd* like to see is some sort of program that really illustrates how off
these calculations really are. I don't think it's as bad as IBM says, but I
don't think it's as infrequent as Intel says either. 🙂
Steve
"What *I'd* like to see is some sort of program that really illustrates how off
these calculations really are."
No such thing, Steve. The nature and extent of the error depends entirely on
the mathematical equation being evaluated. So it can be off just a hair in some
calculations and truly out to lunch in others. The Wall St. Journal ran one
example last week of a legitimate equation that would evaluate to zero –
erroneously.
Jerry
It was funny when Intel said that no one needed divisional math
accuracy after the fourth decimal and then all of the researchers at
NASA, etc… stood up and said, "Do you want to be responsible for
the next Shuttle crash where we use Pentiums to do critical
computations."
Also, IBM recently stopped production of Pentium based systems. They
had found that Intel's portrayal of the occurences of mistakes was
very much underestimated. The internet is now filled with samples of
'simple' calculations that will illustrate errors. There was even a
spreadsheet that someone posted that consistantly creates errors.
The bottom line is that it was a MAJOR flaw and that many more people
are affected than what Intel would like you to believe.
"The bottom line is that it was a MAJOR flaw and that many more people
are affected than what Intel would like you to believe."
Since I own a recently purchased Pentium, I have followed this whole
story pretty closely. All the tests I've seen are pretty far from
what I do in my daily work – and most used contrived numbers. The
example published in The Wall St. Journal last week is illustrative
of this.
Based on what I've seen, I wouldn't personally consider the flaw to
be "major" or for that matter even significant to the vast majority
of Pentium users, including me.
I have however already requested an error-free Pentium replacement.
What blows me away though is Intel's total deadhead attitude in
dealing with this – and how a bunch of people who aren't even very
well acquainted with the actual floating point problem have turned it
into a major news item.
"Also, IBM recently stopped production of Pentium based systems."
IBM has their own axe to grind with Intel and competitive reasons
for making such a decision. IBM produces 486's under license and the
economic impact of their suspending Pentium production for a few
months is minimal.
From a publicity standpoint, it was a smart business move for IBM –
but not necessarily a reflection of the seriousness of the problem.
"The internet is now filled with samples of 'simple' calculations
that will illustrate errors."
I don't doubt it. The five missing entries in the floating point
table have been published by Intel and a reasonably bright eigth
grader could construct all the erroneous calculations you'd ever hope
to see.
The question is could they have done it without knowing precisely
which table entries were missing? I kind of doubt it. Consider that
it took a professional mathematician some time and some pretty
esoteric calculating to discover the flaw in the first place – and
Intel, much to their discredit, knew about it before that and kept
silent.
" . . . and then all of the researchers at NASA, etc… stood up and
said, "Do you want to be responsible for the next Shuttle crash where
we use Pentiums to do critical computations.""
Somehow I seem to have missed that in all the reports I've read. Do
you have a citation to that? Or is just something you've heard? I
would imagine that a story like that would be bigger news than IBM's
announcement.
In short, my very own personal opinion is that it's difficult to consider the
floating point problem a "MAJOR" flaw. But all the "instant experts" on
floating point calculations and their opinions are certainly worth a chuckle or
two – and the Pentium jokes floating around are absolutely priceless!
Jerry
>What blows me away though is Intel's total deadhead attitude in dealing
HEY!!!
Stop defaming us Dead Heads.
>Intel, much to their discredit, knew about it before that and kept silent
That is the truly horrible part of all this.
>Do you have a citation to that?
Several organizations which perform safety critical mathematically complex
computation have taken all of their Pentium systems off line. I heard a report
on the news last night that mentioned a few, I seem to recall NASA and JPL as
thrown in, though I don't know if it was all of NASA period.
"Several organizations which perform safety critical mathematically complex
computation have taken all of their Pentium systems off line."
Yeah, that I heard of and think prudent, but I hadn't heard anything about
*all* of anyone's engineers, blah, blah, blah. With stories like this, I think
it's important that accuracy be maintained.
"Stop defaming us Dead Heads."
You are correct. I amend my previous statement to read "Intel was just plain
stupid . . ." There is no reason a perfectly fine music group or its devoteee
legions should be smeared because of this. <G>
Jerry
According to an article in my local paper (reprinted from the New
Zork Times) a major NYC bank claims to have discovered calculation
errors it attributes to the Pentium. The bank spokesman who didn't
want to identify himself or the bank says their testing found the
innacuracies could occur more frequently than intel says.
The article goes on to say that scientists at Brrokhaven labs found
a miscalculation in calcing the impact of colliding subatomic
particles. A multimedia company in Long Island sent back two Pentium
machines after a miscalculation was found in a graphics program. They
replaced them with Macs.
— I'm waiting until I can afford the PowerPC 🙂
MOS via Autopilot & Amiga
Well, I didn't say the problems didn't exist – just that in context, it is
not a "MAJOR" end of the world type of thing. Information Week published a
survey today of Pentium users. 95% saw no reason to either stop using or buying
Pentiums.
Intel has the new corrected CPU in production and has relented on its
previous (and incredibly stupid) policy of deciding whether or not you need a
new CPU.
So in two months or so, anyone with a flawed Pentium will have one that
works. End of problem.
The whole thing kind of reminds me of the great virus flap a few years back.
Lots of noise and little actual substance.
Jerry
I'm not saying it's a 'major' flap either… it could just be another
media feeding frenzy. It does sound like they're fixing things although
not as fast as some would like. The Times article also said 8 lawsuits had
been filed.
Still you can't blame the Davids for snickering when Goliath stubs his
toe. 😀
-Jim
MOS via Autopilot & Amiga
Jerry,
>The whole thing kind of reminds me of the great virus flap a few years back.
Lots of noise and little >actual substance.
Hmm, isn't that how most political races are won or lost? ;-}
Regards,
John Chang
"Hmm, isn't that how most political races are won or lost?"
Yeah. Computer marketing wars too. <G>
Jerry
Jim,
Oh lovely. I wonder how many Intel Marketing Execs are on suicide watch?
Regards,
John Chang
John,
The same Times article quotes an Intel rep several times. It sounded
like a PR flak to me… but I bet they're worried at intel. More over
their reputation than the bug, which has been corrected.
-Jim
MOS via Autopilot & Amiga
Jim,
Indeed the "bug" is relatively trivial, however, perception is a different
story. Something that Intel did not expect and is a model for other
manufacturers what not to do.
Regards,
John Chang
> The Pentium problems consist of five missing entries in a floating
>point lookup table. Unless the application is doing a huge amount of
>floating point calculations, the "problem" is essentially a non-issue in
>practical terms for the vast majority of users.
True.
> Intel, of course, did a horrible job of dealing with the flaw and it's
>understandable that every user wants a "perfect" Pentium (which is why
>I've already placed a call to Intel's replacement line).
Horrible is an understatement, but essentially true, though it brings
phrases like "I am not a crook" to mind. 🙂
> But there's no reason to overstate the flaw as if it is some crippling
>defect. It isn't.
Actually, I don't think anyone is overstating. This is a serious flaw if
you EVER plan to do any floating point work. There is just no way to
trust the answer now. I or my company will not purchase pentiums untill
this bug (and another timing bug) are worked out. Maybe we don't do
floating point stuff on them right now, but who's to know what that PC is
going to be running later…
Warp 3 — Engage!
Bob Comer Cheyenne, Wyoming U.S.A.
"There is just no way to trust the answer now. I or my company will
not purchase pentiums untill this bug (and another timing bug) are
worked out. Maybe we don't do floating point stuff on them right
now, but who's to know what that PC is going to be running later…"
Actually, all you have to do is run the same problem on a
non-Pentium as a check. In two or three months, Intel should have
caught up on the demand for replacement CPU's. But I sure wouldn't
buy another Pentium without checking to make sure it had the
corrected CPU.
> Actually, all you have to do is run the same problem on a non-Pentium
>as a check.
I know that. It kind of eliminates the need to run it on a pentium,
doesn't it. 🙂
> In two or three months, Intel should have caught up on the demand for
>replacement CPU's. But I sure wouldn't buy another Pentium without
>checking to make sure it had the corrected CPU.
Yep, same here!
Did you hear the latest? Intel said it would replace all pentiums,
no questions asked. Al you have to do is request it. (That's going
to cost Intel a lot of money.)
"Intel said it would replace all pentiums, no
questions asked. Al you have to do is request it. (That's going to cost Intel
a lot of money.)"
And it's exactly what they should have done from the very first.
Jerry
Just heard on CNN Headline News that Intel will offer all Pentium
owners an exchange for a "corrected" Pentium processor, no questions
asked. They reported that it will cost Intel approximately $100
million dollars for the exchange program.
Regards,
John Chang
Yep. That was in The Wall St. Journal and New York Times yesterday.
Guess Dr. Grove finally woke up and smelled the Pentium burning. <G>
Jerry
Jerry,
Hee, hee, hee. I think the boys at Motorola are carefully going through all
the PowerPC chip designs just to make sure they didn't miss anything either!
Happy New Year!
Regards,
John Chang
My local dealer has promised me a new chip at the end of January, free
of charge… I've done the few tests I've come across, and it seems
that my chip _isn't even faulty_!! Frederic Botton from Paris,
France fbotton@cix.compulink.co.uk
Best one I've heard on the Pentium so far….
The new, corrected version is known as the RePentium.
-larry
Larry,
ROFL-PA! Repent! Repent!
Regards,
John Chang
> And it's exactly what they should have done from the very first.
I couldn't agree more!
Warp 3 — Engage!
Bob Comer Cheyenne, Wyoming U.S.A.
"I couldn't agree more!"
Now if only Commodore had done that with the faulty Buster chips. Of course
that didn't make the news. <G>
Jerry
> Now if only Commodore had done that with the faulty Buster chips. Of
>course that didn't make the news. <G>
Nope, it didn't. A different problem though, as it only block expansion,
not falsified answers.
Warp 3 — Engage!
Bob Comer Cheyenne, Wyoming U.S.A.
"Nope, it didn't. A different problem though, as it only block expansion, not
falsified answers."
Still keeps the machine from being used for all its intended purposes, Bob.
But, gee, I suppose I could sue Commodore. <G>
Jerry
> Still keeps the machine from being used for all its intended purposes,
>Bob.
I agree it should have been fixed, but it was a problem I could live with.
(Until I needed to put another card in!) I can see how it would have been
a killer for most A4000 setups though.
> But, gee, I suppose I could sue Commodore. <G>
:/ (I don't know whether to laugh or to cry.)
Anyhow, I think the whole Intel thing is HILARIOUS! Intel's got to
start thinking their a consumer company, or this kind of thing is
going to kill them.
Warp 3 — Engage!
Bob Comer Cheyenne, Wyoming U.S.A.
"Intel's got to start thinking their a consumer company, or this kind of thing
is going to kill them."
Yep, although thinking like a consumer company might kill them just as
easily.
Jerry
Bob,
If you just use Excel on any of your Pentiums, for any divide operations,
you are using floating point divides.
We've got a few people around here double checking thier work on '486
machines just to be sure. Checking calculations on another machine
(Jerry), is not an acceptable solution.
Our company had already demanded replacements before Intel totally backed
down.
To everyone else as well, all my prejudices towards the Amiga aside…
I've never seen an IBM mainframe or midrange computer perform a
calculation that poorly (or any other companies).
It is a 'VERY BIG' deal when you are dealing with millions of accounts
and billions of dollars. It is a 'VERY BIG' deal. Perhaps not the
home user or game player, but for big corporations it's a different
story. Thank goodness we haven't put too many mission critical
applications on PC's, yet.
Robb
>If you just use Excel on any of your Pentiums, for any divide operations,
>you are using floating point divides.
Yep!
>Our company had already demanded replacements before Intel totally backed
>down.
We only have 1 pentium machine in house, so it shouldn't be to hard to get
a couple of new pentiums. (Dual pentium machine.) We haven't upgrade any
of our 486's yet, and now we don't plan to until the pentium is fixed.
>To everyone else as well, all my prejudices towards the Amiga aside…
>I've never seen an IBM mainframe or midrange computer perform a
>calculation that poorly (or any other companies).
I agree, it's totally inexcusable in the computer world.
>Thank goodness we haven't put too many mission critical applications on
>PC's, yet.
Amen! (We haven't either.)
Warp 3 — Engage!
Bob Comer Cheyenne, Wyoming U.S.A.
"Thank
goodness we haven't put too many mission critical applications on PC's,
yet."
Out of curiousity, what kind of work do you do that involves floating point
calculations?
And my suggestion of checking out the math on another machine applied only to
determining if the machine had a defective Pentium that was affecting
calculaton, not for production processes. That part of it is partially mooted
by Intel's final cave-in/
Jerry
Jerry,
>>"Thank goodness we haven't put too many mission critical applications on
>>PC's, yet."
>>
>>Out of curiousity, what kind of work do you do that involves floating
>>point calculations?
A week later…I've been too busy with the holidays here, Happy Season to
you!
As I tried to say, not much, on PCs. I'm really not sure how extensive our
use of floating point is on the mainframes. I'm sure that a lot of systems
use it.
We are doing more and more mission critical stuff on PCs though. Like one
project I'm working on that involves payment processing and reject
re-entry. I don't know whether or not the image processing systems use
floating point math to do image recognition or not, but I don't think we
should have to worry about it. Maybe the next release?
Do you want the handwriting on your loan payment check processed by a
Pentium?
How comfortable would you feel if you went into one of the top 5 banks in
the country for a multi million dollar loan and the guy at the desk was
using Excel on a Pentium to do some preliminary calculations,
amortizations, etc?
I'm sure you'd have the numbers checked on your own in that case even if
it wasn't a Pentium (as I would), but I think you can see my point?
Robb
Rob:
None of the applications you mention would need to use floating-point math.
There's no argument between us here: who wants a CPU that you know is
defective? I sure don't. Intel finally did the logical thing, so we can all
have nice, new Pentiums. <G>
Jerry
Jerry,
Excel does use floating point math, not that it always needs to. So you
never know.
But, if you know a machine has a Pentium and it's a major part of some
transaction you are doing and you don't have control over that machine,
how will you ever know if the transaction is really right? Especially if
it involves alot of calculations? How will you know that the company
replaced the Pentium with a properly functioning one?
Bad move by Intel to sell the CPU knowing that it had a problem and
potentially bad for all the big customers that have bought them because
thier customers will be paranoid if they know that someone that is holding
thier money uses them.
Remember the great thing about this distributed computing thing is that
the clients have control over thier machines. There is limited centralized
change control. Any desktop could have a problem and the user might not be
concerned about it and the bank wouldn't know about it.
Robb
PS. Good luck with your bank accounts everyone.
I think we're beating a dead horse here: Intel will replace any Pentium, no
questions asked. I'd sure be wary of purchasing any used Pentium machines
because, as you correctly point out, you'd have to determine whether or not the
machine had a defective CPU.
"Good luck with your bank accounts everyone."
That's really gratuitous – and probably overstates the possible inaccuracies
of the defective Pentium. In any event, you do balance your own accounts, don't
you? <G>
Jerry
Jerry,
Close to beating a dead horse, but you only have at most (I'm guessing) a
handfull of Pentiums. We have hundreds (if not more) scattered all over in
various departments across several states that may or may not get them
replaced.
There is no way to guarantee that one of them won't hit us one day.
>>In any event, you do balance your own accounts, don't you? <G>
I balance my own, of course, but do you really realize how many
millions of transactions and trillions of CPU cycles a corporation
like ours does in a day? There are a lot of checks and balances on
the mainframe side, but not on the workstation side. There's alot of
room for miscalculations that may not be detected right away.
"Close to beating a dead horse, but you only have at most (I'm guessing) a
handfull of Pentiums."
Not even a handful – just one – which does, of course, have a defective CPU.
I've already called for a replacement.
"I balance my own, of course, but do you really realize how many millions
of transactions and trillions of CPU cycles a corporation like ours does
in a day?"
If it's all integer arithmetic, they shouldn't be affected. Anyway,
I'm not really arguing the point, other than to suggest that the
impact of the problem was vastly overblown and, at this point,
totally academic.
Jerry,
One Pentium could be considered a handful.
Let's send this horse to the rendering plant…
Even though I think that Intel made a bad move in producing and selling a
CPU they knew was bad and then made a bigger mistake by not offering a
free replacement to anyone with one (and I love it!) I don't think it's
that big a deal. Although, I don't feel that there is any excuse for it.
I think that they really have done a lot of damage just because of the
media hype that ensued due to a few poor decisions.
My main point was that the sudden media surge will leave a mark in peoples
minds and in the future they may think about it when it comes to someone
doing work for them on a Pentium machine. It may also cast a shadow of
doubt on that company. Whether or not that software is doing FDIVs.
Then again you really never know how the program is coded.
Robb
>> "Good luck with your bank accounts everyone."
> That's really gratuitous …
It is not. It accurately states the case; that you cannot, without
some special precautions, possibly _KNOW_ that the results of an FDIV
are correct. That is totally unacceptable behaviour from a CPU chip,
and Intel (finally) finally knows and admits it.
-larry
Robert,
Picture one of the Pentium commercials, all of a sudden a "divide by zero"
error appears on the screen and then the "Energizer Bunny" comes across the
screen with the announcer saying, "It keeps going and going and going…" ;-}
Regards,
John Chang
GOOD ONE JOHN 🙂
And John keeps going and going…
Ed,
>And John keeps going and going..
Yup! And trying to avoid the wrath of DJ's rubber chicken! (Enter main theme
from Star Wars: Empire Strikes Back) ;-}
Regards,
John Chang