#Pentium math
21 messages in this thread
They may not be shipping in volume, but they are shipping. I was just bouncing
over in the NewTek group, and there was discussion of how the Pentium bug would
affect Lightwave. Frankly, there aren't many programs other than Lightwave
that I'm worried about, since developers try to avoid floating point for
performance. Anyway, one of the people there said that they had just purchased
five new Pentiums, and none of them had the defective chip. Also, one of the
NewTek guys chimed in to say that the algorithms Lightwave uses won't hit the
Pentium math bug.
If Lightwave's not going to hit it, I'm not that worried about the whole thing.
Of course, I will be checking with Micron when I buy my machine to see what
their policy is to replace the chip. While it may be an incredibly rare
problem, I don't particularly won't to fork over 3,600 for a machine whose
processor has "an incredibly rare problem." 🙂
-R
"..since developers try to avoid floating point for performance".
Interestingly enough, (I haven't checked this because I don't have access to a
Pentium), in one of the graphics newsgroups I read, there was a thread that was
talking about integer math vs. floating point math for graphics programming.
Several people stated they were surprised that the integer math computations
for their 3D applications didn't buy them a lot when they ran on the Pentium,
because the floating point speed was so good. I don't claim this is true, but
as I said, several people piped up and said the same thing.
Wow! That's amazing! Avoiding the floating point seems to be the mantra of
all the devs I know.
-R
And just when Intel puts out an FPU that kicks butt, the shoe gets stuck
in the crack …
DJ
On AP from Queens, the thrill and excitement of the Big Apple
Where Santa's always got a brand new bag …
Steve,
This is generally true for Pentium, MIPS & Alpha processors. In particular
floating point on the RISC chips is astonishingly quick.
–Ben
…via AutoPilot
> they were surprised that the integer math computations for their 3D
> applications didn't buy them a lot when they ran on the Pentium, …
When I make wild guesses about an answer, instead of being accurate, I can do
it just as fast in integer or float. <chortle>
-larry
It is interesting to hear Intel's spin on this problem. According to Andy
Grove, the frequency of the bug is once in nine billion floating point divides.
However, according to the Intel Pentium Reference, 1 FDIV takes 39 cycles. At
90 MHz, this roughly comes to a floating point error once every 65 minutes. At
100 MHz, it is 58.5 minutes. In other words, what Intel is saying is that the
error occurs once every hour.
..Andy from Harding NJ
This morning's news says that IBM has suspended shipping of Pentium-based
systems until the floating-point error question has been cleared up. It was
not clear from the (radio) news report as to whether this was IBM Canada only
or IBM worldwide, but I suspect the latter.
The news report repeated Intel's claim of "only one error per 27 years",
which gives me an inspiration:
I'll waltz over to a friend who has a Pentium system, and slip in the
calculation [ (4195835/3145727) * 3145727 -1 ] … doing it TWICE. I'll then
assure my friend that, according the Intel, he's now assured of 54 years of
error free calculation. 🙂
–Jim
Jim,
It's IBM worldwide.
Don
Speaking of Pentium math, a friend of mine mailed the following calculation:
4.999999 / 14.999999 * 3.000001
Which should be about: 1.00000019
Word is the Pentium thinks it's less than one. Whoops.
(Haven't tried it myself, yet).
Steve
Pentium – from the Latin meaning "Almost Five".
Gosh, another specific example of a Pentium math failure.
Congratulate your friend for his effort in spending 27 years finding this
one. 🙂
[ ..or is it 27,000? Press reports are inconsistent.. ].
–Jim
<grin!>
Someone sent me a fake press release today that said Intel stock was splitting
3 to 1.98347553. 🙂
Har! 10.0!
-larry
> <grin!>
>
> Someone sent me a fake press release today that said Intel stock was
splitting
> 3 to 1.98347553. 🙂
More facts I've dug up…..
The Democrats are still shaken over their unexpected losses during the Nov
mid-term elections. "I just don't understand!", exclaimed Democratic pollster
Robert Dudem. "We took advantage of the Orange County Investment Fund's offer
of unlimited use of their recently upgraded Pentiums". He went on to explain
that "The numbers kept predicting an overwhelming win. We were seeing 130.234
out of 100 voters supporting us". He expressed sadness that those numbers did
not materialize on election day, but fully expected to make good in '96.
Experts agree that the PowerPC line of computer chips developed by IBM,
Motorola, and Apple were not going to be a match for Intel's soon to be
released P6 CPU. At a press conference held at the San Jose Hilton yesterday,
Intels chairman, Andy Grove, explained what set the P6 apart. "This is the
first CPU ever desiged 100% by it's predecessor. We used the amazing power of
the Pentium to design the advanced logic required by the P6". Expected
performance figures were also released. Mr Grove indicated that Pentium
simulations predicted over 200 SPECmark for the P6 without any power applied!
At 3.3v, Pentium simulations indicate 250 SPECint. At 5v, 450 SPECint is
expected. The bombshell however was Intel's claim that at 110 volts, over 1000
SPECint was possible! "This puts PowerPC out
of the game!" indicates industry analyst Peter Snugman. "Our Pentium validated
projections indicate that Intel can sell P6's at a loss and make up the money
by sheer volume." Asked to back this up, he answered "Do the math yourself!"
——- Andrew
As usual, a direct hit to the funny bone!
Paul – wearing my black construction paper funeral suit
Andrew,
ROFL-PA! OK! OK! Stop it! You're going to kill me! Hey Tim Walsh of
AmigaWorld! If you're reading this, you gotta print this stuff in the February
or March issue!
Regards,
John Chang
P.S. Ohh! My stomach hurts! ;-}
"Ohh! My stomach hurts!" – what do you expect after eating rubber chicken
Buffalo wings???
Speaking of AmigaWorld, what's the scoop with Amazing Computing? I
haven't received anything in a while. Or am I getting paranoid?
DJ
On AP from Queens, the thrill and excitement of the Big Apple
Where the ice cream truck plays Helter Skelter …
You are experiencing the wonders of the US mail system, unless you let your
subscription run out. I got an issue a couple of weeks ago.
-Karl
DJ:
According to a post on c.s.amiga.misc Amazing missed their print
deadline for the December issue because they had hoped to scoop the news
regarding the finalizing of the liquidation. One source says they'll
be mailing it late, another says it'll be skipped entirely. We shall
see what we shall see.
Stuart
(who STILL wants his 4000T by March!)
DJ,
>"Ohh! My stomach hurts!" – what do you expect after eating rubber chicken
>Buffalo wings???
Touche! ;-}
As far as I know, everything at Amazing is fine.
Regards,
John Chang
Can you but love it? Who says there is no God? A great diversion while
we wait … and wait …
DJ
On AP from Queens, the thrill and excitement of the Big Apple
Where the ice cream truck plays Helter Skelter …