

December, 1994



        Intel PENTIUM flaw has computer world abuzz
        ===========================================


        The recent revelation, and subsequent admission by

Intel, that the Pentium chip has a serious flaw relating to

floating point math calculations has the statistical and

simulation sector worried. Intel said the error occurs

approximately once in every nine (9) billion divides.

However, a few thousand number pairs already have shown

to cause errors.



THE PROBLEM: A rounding error occurs in some floating-point divide
             operations, affecting numbers in the 4th thru 19th
             positions after the decimal point.


TESTING
PROCEDURE:   Perform this operation:

             (4,195,835/3,145,727)*3,145,727-4,195,835=

             The result should be zero; a flawed Pentium processor
             will return a NON-ZERO answer.


APPLICABLE
CHIPS:       All Pentium chips currently on the market. Intel plans
             to ship reworked Pentium chips within the next month to
             users with "verifiable" problems.
              
             
AREA OF
CONCERN:     Users heavily involved in financial applications,
             mathematically-intensive simulation or statistical
             analysis are deemed at risk. Any computer operations
             that perform millions of high-precision divide operations
             daily are the prime areas of concern.

             The following are NOT considered affected by this flaw:

             1) Users who primarily do spreadsheets, word processing
                and conventional database applications.

             2) Those users who use the Pentium for file servers.

             3) Users who never use the math coprocessor.

POINT OF
CONTACT:     Intel Corp. toll free number is 1-800-628-8686.


POSSIBLE
SOFTWARE 
BUG-FIX:     Compaq Computer Corp. has proposed a stopgap software
             solution for the interim. They plan to distribute a
             software utility that will disable the floating-point
             capabilities, creating, in effect, a Pentium SX, which
             is estimated to slow floating-point operations by a
             factor of about ten.

BACKGROUND:  It should be noted that the first 486 chips similarly
             had floating-point errors (bugs); Intel eliminated the
             math coprocessor, resulting in the 486SX.


//////////////////////////////////////////////////////////////////

Another in a series of FREEWARE files brought to you by PUBNETIX.
COPYRIGHT 1994 PUBNETIX. ALL RIGHTS RESERVED. This file may be
freely copied/distributed on BBS provided file is not altered and
no fee is charged for its dissemination.

//////////////////////////////////////////////////////////////////
