United States Nuclear Regulatory Commission - Protecting People and the Environment

ACCESSION #: 9505050145

                        Boeing Computer Services
                             P.O. Box 24346
                         Seattle, WA 98124-0346

April 25, 1995
G-1151-RSO-95-186

Document Control Desk
United States Nuclear Regulatory Commission
Washington, D.C. 20555

BOEING

Reference:     a)   Boeing Letter G-1151-RSO-92-365 dated August 31,
                    1992; R. S. Orr to the NRC Operations Center

               b)   NRC Letter Docket No. 99901227 dated August 12, 1992;
                    L. J. Norrholm to R. S. Orr; Subject: Response to 10
                    CFR 21 Inquiry

Dear Sir or Madam:

In accordance with the reference correspondence and 10 CFR 21, Boeing is
sending the NRC the attached error notice(s) received from our former
software suppliers.  Because of unknown current addresses, the following
former customers were not notified:

     Reactor Controls, Inc.

     Echo Energy Consultants, Inc.

     Nuclear Applications and Systems Analysis Company (Japan)

     Nuclear Power Services

Error notices have been sent to our other former customers.

Very truly yours,

R. S. Orr
Nuclear Administrator
G-1151 M/S 7A-33
(206) 865-6248

Attachment(s): GTICES Program Report Form No. 95.10
               ANSYS Class3 Error Reports 95-12, 95-13, 95-14 and 95-15,
               Reissued Class3 Error Reports 94-01 R1, 94-68 R1, and
               95-10 R1;
               ANSYS QA Notice QA95-04

Georgia Tech             Computer Aided Structural Engineering Center
                         Georgia Institute of Technology

                         School of Civil & Environmental Engineering
                         Atlanta, Georgia 30332-0355
                         USA
                         404o894o2260
                         404o894o8014 FAX

April 18, 1995

Attention: Nuclear Administrator
Boeing Computer Services
P.O. Box 24346, M/S 7A-33
Seattle, Washington 98124-0346

RE:  GT STRUDL

Dear Sir or Madam:

Enclosed please find a copy of GTICES PROGRAM REPORT FORM No. 95.10 and a
VENDOR ACKNOWLEDGEMENT FORM.  Please sign and return the VENDOR
ACKNOWLEDGEMENT FORM to acknowledge receipt of the GTICES Program Report.

Thank you for reviewing the Program Report and for returning the
Acknowledgement Form.

Best regards,
CASE Center

Catherine Lee
Configuration Control Manager

CL/jhb
Enclosures

                      GTSTRUDL Program Report Form

                                             GPRF No.: 95.10

                                                  DATE: April 17, 1995

FROM:     Computer-Aided Structural Engineering Center
          Georgia Institute of Technology
          Atlanta, Georgia 30332-0355

SEVERITY LEVEL:

     URGENT         Problem results in incorrect answers which may not be
                    apparent or job aborts and cannot be recovered within
                    the session or job.

x    SERIOUS        Problem results in incorrect answers which are
                    obvious or problem prevents completion of a
                    particular user's task.

     MINOR          Problem can be worked around or problem poses high
                    frustration factor.

     INFORMATIVE    Documentation error, program usage tip, user
                    inconveniences.

Date Problem Confirmed   April 17, 1995

Date Notification Sent   April 19, 1995

Computers  VAX

Operating System    VMS

Version  All VAX versions

Target Release for Correction 96.01

                                             Senior Software Engineer
Signature                                    Title
R & D Division

Hamid Zand                                   April 17, 1995
Typed or Printed Name                        Date of Signature

Signature                                    Title
Professional Services Division

Typed or Printed Name                        Date of Signature

                                                       Rev. 2.2

                      GTSTRUDL Program Report Form
                               (Continued)

                                        GPRF No.: 95.10

                                             DATE: April 17, 1995
DESCRIPTION:

                             JOINT RELEASES

An error message may not be printed when two or more blank spaces are
typed after the misspelled word RELEASES (RELEASES ) in the JOINT
RELEASES command on the VAX/VMS version of GTSTRUDL.
     Following is an example of misspelled word RELEASES followed by two
blank spaces (RELEASES ).

STRUDL

***  G T S T R U D L ***
     RELEASE DATE             VERSION        COMPLETION NO.
     AUGUST 1993             9301VAX          3062

***  ACTIVE UNITS - LENGTH    WEIGHT    ANGLE     TEMPERATURE    TIME
***  ASSUMED TO BE       INCH POUND     RADIAN    FAHRENHEIT  SECOND

UNITS KIPS FEET
JOINT COORDINATE
1    0.0  0.0
2    0.0  10
3.   10   10
4.   10   0.0

STATUS SUPPORT 1 4
JOINT RELEASES
1 4 MOM X MOM Y MOM Z
%GTCMI-W-INVCOMMND, Invalid command /1/

TYPE PLANE FRAME
MEMBER INCIDENCE
1    1    2
2    2    3
3    4    3
*FIN $ EOF

                                                       Rev. 2.2

                      GTSTRUDL Program Report Form
                               (Continued)

                                             GPRF No.: 95.10

                                                  DATE: April 17, 1995

Following is the same example when the two blank spaces are deleted after
the misspelled word RELEASES (RELEASES)

STRUDL

*** G T S T R U D L ***
     RELEASE DATE        VERSION        COMPLETION NO.
     AUGUST 1993         9301VAX        3062

***** ACTIVE UNITS - LENGTH WEIGHT ANGLE TEMPERATURE TIME
***** ASSUMED TO BE INCH POUND RADIAN FAHRENHEIT SECOND

UNITS KIPS FEET
JOINT COORDINATE
1    0.0  0.0
2    0.0  10
3    10   10
4    10   0.0

STATUS SUPPORT 1  4
JOINT RELEASES
%GTCMI-W-COMNOTPRO, Command not completely processed.
RELEASES
 1   4    MOM  X    MOM  Y    MOM  Z
%GTCMI-W-INVCOMMND, Invalid command /1/

TYPE PLANE FRAME
MEMBER INCIDENCE
1    1    2
2    2    3
3    4    3
*FIN $ EOF

Applicable Sections of the Documentation:

     JOINT RELEASES -    Section 2.1.7.2, Volume 1,
                         GTSTRUDL User's Manual

                                                       Rev. 2.2

ANSYS (Registered Trademark)

                         ANSYS, Inc.
                         201 Johnson Road         Telephone 412.746.3304
                         Houston, PA 15342-1300   Facsimile 412.746.9494

                             April 12, 1995

Dear Class3 Error Recipient:

     Enclosed you will find ANSYS Class3 Error Reports 95-12, 95-13,
95-14 and 95-15.  Along with these reports are reissued Class3 Error
Reports 94-01 R1, 94-68 R1, 95-10 R1 and ANSYS QA Notice QA95-04.

     Class3 Error Report 94-01 was revised to indicate that Component
Products are also affected by this error.

     Class3 Error Report 94-68 was revised to include nonlinear effects.

     Class3 Error Report 95-10 was revised to indicate that only base
excitation spectrum analysis has this error.

     ANSYS QA Notice QA95-04 describes our findings related to the Intel
Pentium Floating Point unit flaw and its impact on ANSYS analyses.

                                        Sincerely,

                                        ANSYS, Inc.

                                        Prashant Ambe
                                        Quality Assurance Manager

    ANSYS (Registered Trademark) is a registered trademark of SAS IP

                  ANSYS Revision Identifier Description

     ANSYS revision identifiers consist of a major revision level, a
minor revision level, a correction level, and occasionally a special
version level.  An example of how this is constructed is shown below:

Figure "5.0A-10" omitted.

     Major revision level changes indicate that new features have been
added to the program and that some level of program architecture change
and/or file structure has occurred.  Minor revision level changes also
indicate that new features have been added to the program, but files are
upwardly compatible.  All known error fixes are included in both minor
and major revisions.  Changes to the correction level indicate that it is
primarily an error correction release.  Special version identifiers
indicate that one or more additional minor changes have been made to the
program, normally to circumvent an error.  Special versions are not
general releases to all ANSYS licensees, since they typically represent
errors occurring only on one system, a subset of our customers who have
specific graphics devices, etc.

     The ANSYS revision identifier(s) shown under "corrected in" on the
front side of this Class3 Error Report indicates the first possible
revision that could contain the correction.  A major program change
needed to fix an error can dictate that the next minor or major revision
will contain the fix rather than the next correction level.  For example,
when errors were being reported while Rev. 4.3A was the latest production
version, most Class3 error reports indicated that 4.3B was the "corrected
in" revision.  Others requiring significant code restructuring were
reported as fixed in 4.4.  Rev. 4.3B was never released, but Rev. 4.4
contained all error corrections noted as fixed in 4.3B.

     An identifier indicated under "corrected in" does no guarantee that
a general release of that revision of ANSYS will occur.  It does indicate
that the correction is known and implemented in the coding that would be
part of that general release.

                     Equivalent Product Identifiers

The ANSYS family of component products occasionally undergoes name
changes between revisions and/or changes in the functionality of derived
products (such as ANSYS-PC/LINEAR).  To minimize the potential for
confusion in these areas, the important product name equivalences
(similar program functionality and error content) are listed below.

     ANSYS/ED  contains all errors shown for      the full ANSYS product,
                                                  starting at Rev. 5.0,
                                                  and beyond, unless
                                                  otherwise noted.

     PREPPOST  contains relevant errors           the full ANSYS product,
               shown for                          for included pre- and
                                                  postprocessing
                                                  functionality

     ANSYS Component Products is an               All component products
           inclusive term for                     that were issued at
                                                  the stated revision
                                                  level(s)

     PC/LINEAR   is equivalent to                 WS/LINEAR at Rev.4.4A,
                                             and  ANSYS/LinearPlus
                                                  starting at Rev. 5.0A.

     PC/THERMAL  is equivalent to                 WS/THERMAL at Rev.4.4A,
                                             and  ANSYS/THERMAL
                                                  starting at Rev. 5.0A.

     PC/MAGNETIC is equivalent to                 ANSYS/Emag starting at
                                                  Rev. 5.0A

            ANSYS (Registered Trademark) CLASS3 ERROR REPORT

                                                  ERROR NO: 95-12

KEYWORDS:      SPECTRUM       MODE COMBINATION    GROUPING METHOD

DESCRIPTION OF ERROR:

     In the grouping method of mode combination (MCOMB,GRP) in spectrum
analysis (ANTYPE, SPECTRUM) there are two errors as follows:

1.   if isolated and closely-spaced modes are required to be combined,
     isolated modes are erroneously discarded, and

2.   the negative values of the mode coefficients for the closely-spaced
     modes are not turned into positive values before writing to the file
     .mcom.

FIRST INCORRECT VERSION(S):*_/          CORRECTED IN:*_/

Rev. 5.0                                Rev. 5.2
ANSYS/LinearPlus Rev. 5.0A              ANSYS/LinearP]us Rev. 5.2

SUGGESTED USER ACTION FOR RUNNING ON UNCORRECTED VERSION:

     Edit an incorrect file.mcom (containing the ANSYS commands which
specify the combinations) to reflect a correct grouping method mode
combination for the particular problem at hand and use it in POST1.  Note
that this requires familiarity with the POST1 load case operation
commands.

COMMENTS:

     Users may use one of the other mode combination methods available in
the ANSYS program.

AUTHOR/CORRECTOR:                       DATE:     4/10/95
                    Ashraf Ali

REVIEWED BY QA:                         DATE:     4/10/95

REVIEWED BY QA:                         DATE:     4/10/95
                    Prashant Ambe

APPROVAL:                               DATE:     4/10/95
                    John A. Swanson

*_/ If a product name is not included in the "first incorrect version",
the full ANSYS program is implied.  For products not listed, this error
does not apply, but see the reverse side for equivalent product
designations.

Unless noted otherwise, this error report also applies to all revisions
after the first incorrect one and prior to the corrected revision.  All
revisions after "corrected in" are corrected.  ManuaL corrections are
included in on-line documentation as appropriate.  Please see the reverse
side of this sheet for additional information on ANSYS revision
identifiers.

                                                  FORM SASI-QA25
                                                  DEC 19, 1988

            ANSYS (Registered Trademark) CLASS3 ERROR REPORT

                                                  ERROR NO: 95-13

KEYWORDS: ELEM3     ELEM4     ELEM54    STRESS STIFFENING   LARGE
                                                            DEFLECTION

DESCRIPTION OF ERROR:

     Results due to stress stiffening effects (PSTRES,ON or SSTIF,ON) in
elements BEAM3, BEAM4, and BEAM54 are incorrect if large deflection
(NLGEOM,ON) is also active.

FIRST INCORRECT VERSION(S):*_/               CORRECTED IN:*_/

Rev 5.0                                      Rev. 5.1 (BEAM4)
                                             Rev. 5.2 (BEAM3, BEAM54)
ANSYS/LinearPlus Rev. 5.0A                   ANSYS/LinearPlus Rev. 5.2

SUGGESTED USER ACTION FOR RUNNING ON UNCORRECTED VERSION:

               If available, use a beam element that does not have this
error (BEAM23, BEAM24, or BEAM44).

COMMENTS:

     Stress stiffness is not necessary with large deflection, but may
speed convergence.

AUTHOR/CORRECTOR:                            DATE: 4/10/95
                         Cecil R. Rogers

REVIEWED BY QA:                              DATE: 4/10/95
                         Prashant Ambe

APPROVAL:                                    DATE: 4/10/95
                         John A. Swanson

*_/If a product name is not included in the "first incorrect version",
the full ANSYS program is implied.  For products not listed, this error
does not apply, but see the reverse side for equivalent product
designations.

Unless noted otherwise, this error report also applies to all revisions
after the first incorrect one and prior to the corrected revision.  All
revisions after "corrected in" are corrected.  Manual corrections are
included in on-line documentation as appropriate.  Please see the reverse
side of this sheet for additional information on ANSYS revision
identifiers.

                                             FORM SASI-QA25
                                             DEC 19, 1988

            ANSYS (Registered Trademark) CLASS3 ERROR REPORT

                                             ERROR NO: 95-14

KEYWORDS:      ELEM34    ELEM71    TEMPERATURE-DEPENDENT    MATERIAL
                                   PROPERTIES

DESCRIPTION OF ERROR:

               Certain element matrices for LINK34 (convection link) and
MASS71 (lumped thermal mass) are not always updated properly if the
material properties are temperature-dependent and the temperatures are
changing:

     1.   For LINK34, the conductivity matrix may not be updated.

     2.   For MASS71, the specific heat matrix is not updated if the only
          temperature-dependent material property for the element is
          specific heat (C).

FIRST INCORRECT VERSION(S):*_/          CORRECTED IN:*_/

Rev. 5.0                                Rev. 5.2
ANSYS/Thermal Rev. 5.0A                 ANSYS/Thermal Rev. 5.2

SUGGESTED USER ACTION FOR RUNNING ON UNCORRECTED VERSION:

     For error 1, set the Newton-Raphson key (NROPT, FULL) to force
matrix updating.

     For error 2, make density (MP,DENS) slightly temperature-dependent.

COMMENTS:

AUTHOR/CORRECTOR:                            DATE: 4/10/95
                    Cecile R. Rogers

REVIEWED BY QA:                              DATE: 4/10/95
                    Prashant Ambe

APPROVAL:                                    DATE: 4/10/95
                    John A. Swanson

*_/ If a product name is not included in the "first incorrect version",
the full ANSYS program is implied.  For products not listed, this error
does not apply, but see the reverse side for equivalent product
designations.

Unless noted otherwise, this error report also applies to all revisions
after the first incorrect one and prior to the corrected revision.  ALL
revisions after "corrected in" are corrected.  Manual corrections are
included in on-Line documentation as appropriate.  Please see the reverse
side of this sheet for additional information on ANSYS revision
identifiers.

                                                  FORM SASI-QA25
                                                  DEC 19, 1988

            ANSYS (Registered Trademark) CLASS3 ERROR REPORT

                                                  ERROR NO: 95-15

KEYWORDS:      LINK34    ELEM34    FILM COEFFICIENT    SFE

DESCRIPTION OF ERROR:

     The convection film coefficient (HF) is incorrect if input on the
SFE or SF commands for LINK34 (convection link) with KEYOPT(3)=2 and
ramped boundary conditions (KBC=O [default]).

FIRST INCORRECT VERSION(S):*_/          CORRECTED IN:*_/

Rev. 5.0                                Rev. 5.2
ANSYS/Thermal Rev. 5.0A                 ANSYS/Thermal Rev. 5.2

SUGGESTED USER ACTION FOR RUNNING ON UNCORRECTED VERSION:

     Run smaller substeps with stepped boundary conditions (KBC=1) and
only one substep per load step (NSUBST=1).

COMMENTS:

AUTHOR/CORRECTOR:                            DATE: 4/10/95
                    Cecil R. Rogers

REVIEWED BY QA:                              DATE: 4/10/95
                    Prashant Ambe

APPROVAL:                                    DATE: 4/10/95
                    John A. Swanson

*_/ If a product name is not included in the "first incorrect version",
the full ANSYS program is implied.  For products not listed, this error
does not apply, but see the reverse side for equivalent product
designations.

Unless noted otherwise, this error report also applies to all revisions
after the first incorrect one and prior to the corrected revision.  All
revisions after "corrected in" are corrected.  Manual corrections are
included in on-line documentation as appropriate.  Please see the reverse
side of this sheet for additional information on ANSYS revision
identifiers.

                                                  FORM SASI-QA25
                                                  DEC 19, 1988

            ANSYS (Registered Trademark) CLASS3 ERROR REPORT

                                                  ERROR NO: 94-01 R1

KEYWORDS:      APDL      CHARACTER PARAMETERS     SYSTEM    *IF  PC

DESCRIPTION OF ERROR:

     On PC systems only, character parameters or character strings
(enclosed in single quotes) may not produce correct results when used
with the *IF or *ELSEIF commands.

FIRST INCORRECT VERSION(S):*_/               CORRECTED IN:*_/

Rev. 5.0A                                    Rev. 5.1
Component Products Rev. 5.0A                 Component Products Rev. 5.2

SUGGESTED USER ACTION FOR RUNNING ON UNCORRECTED VERSION:

COMMENTS:

AUTHOR/CORRECTOR:                            DATE: 4/10/95
                    Christopher W. Aiken

REVIEWED BY QA:                              DATE: 4/10/95
                    Prashant Ambe

APPROVAL:                                    DATE: 4/10/95
                    John A. Swanson

*_/ If a product name is not included in the "first incorrect version",
the full ANSYS program is implied.  For products not listed, this error
does not apply, but see the reverse side for equivalent product
designations.

Unless noted otherwise, this error report also applies to all revisions
after the first incorrect one and prior to the corrected revision.  All
revisions after "corrected in" are corrected.  Manual corrections are
included in on-line documentation as appropriate.  Please see the reverse
side of this sheet for additional information on ANSYS revision
identifiers.

                                             FORM SASI-QA25
                                             DEC 19, 1988

            ANSYS (Registered Trademark) CLASS3 ERROR REPORT

                                             ERROR NO: 94-68 R1

KEYWORDS: BEAM44    ELEM44         STIFFNESS      RELEASES

DESCRIPTION OF ERROR:

     There are two errors when using BEAM44 (3-D tapered unsymmetric beam
element) with stiffness releases.

1.   All results are in error for load steps, substeps, and iterations in
     which the element stiffness matrix is being re-used.

     Stiffness matrices are generally reused unless large deflections are
     active, or the materials or real constants are changing.  See
     Section 13.3 of the Theory Manual.

2.   All results are incorrect if any of the following are active:

     a.   Large deflection (NLGEOM,ON)
  or b.   Stress Stiffening (SSTIF,ON)
  or c.   Newton-Raphson (NROPT command).

FIRST INCORRECT VERSION(S):*_/          CORRECTED IN:*_/

Rev. 4.4                                Rev. 5.2
ANSYS/LinearPlus Rev.  4.4              ANSYS/LinearPlus Rev. 5.2

SUGGESTED USER ACTION FOR RUNNING ON UNCORRECTED VERSION:

Error 1.  Use KUSE,-1 to force reformulation of all element stiffness
          matrices.

Error 2.  Use coincident coupled nodes (CP command) to induce the
          stiffness release where required.

COMMENTS:

     On some computer systems, he e errors will cause an abort.

     Stiffness releases are accessed with KEYOPT(7) and KEYOPT(8) or, if
using pre-Rev. 5.0 input, with KEYOPT(3) and KEYOPT(4).

     For error 2, if using pre-Rev. 5.0 input:

     a.   Large deflection is activated by the KAY, 6, 1 command.
     b.   Stress stiffening is activated by the KAY, 8, 1 command.
     c.   Newton-Raphson is activated by KAY, 9, n (n = 1, 2, or 3).

AUTHOR/CORRECTOR:                            DATE: 4/10/95
                    Cecil Rogers

REVIEWED BY QA:                              DATE: 4/10/95
                    Prashant Ambe

APPROVAL:                                    DATE: 4/10/95
                    John A. Swanson

*_/ If a product name is not included in the "first incorrect version",
the full ANSYS program is implied.  For products not listed, this error
does not apply, but see the reverse side for equivalent product
designations.

Unless noted otherwise, this error report also applies to all revisions
after the first incorrect one and prior to the corrected revision.  All
revisions after "corrected in" are corrected.  Manual corrections are
included in on-line documentation as appropriate.  Please see the reverse
side of this sheet for additional information on ANSYS revision
identifiers.

                                        FORM SASI-QA25 12/19/88

            ANSYS (Registered Trademark) CLASS3 ERROR REPORT

                                                  ERROR NO: 95-10 R1

KEYWORDS: SPECTRUM PFACT      ROTATED NODES  COUPLED NODES 
CONSTRAINT
                                                            EQUATIONS

DESCRIPTION OF ERROR:

     The base excitation spectrum analysis (ANTYPE, SPECTR and PFACT,
TBLNO, BASE) gives incorrect results if any of the base nodes are either:

     1.   rotated (NROTAT command), with non-zero angle specified.

or   2.   contained in a coupled node set command), (CP, etc. command).

or   3.   used in a constraint equation (CE,CERIG,CEINTF, etc. commands).

FIRST INCORRECT VERSION(S):*_/               CORRECTED IN:*_/

Rev. 5.0                                     Rev. 5.2

SUGGESTED USER ACTION FOR RUNNING ON UNCORRECTED VERSION:

COMMENTS:

AUTHOR/CORRECTOR:                            DATE: 4/10/95
                    Cecil R. Rogers

REVIEWED BY QA:                              DATE: 4/10/95
                    Prashant Ambe

APPROVAL:                                    DATE: 4/10/95
                    John A. Swanson

*_/ If a product name is not included in the "first incorrect version",
the full ANSYS program is implied.  For products not listed, this error
does not apply, but see the reverse side for equivalent product
designations.

Unless noted otherwise, this error report also applies to all revisions
after the first incorrect one and prior to the corrected revision.  All
revisions after "corrected in" are corrected.  Manual corrections are
included in on-line documentation as appropriate.  Please see the reverse
side of this sheet for additional information on ANSYS revision
identifiers.

                                                  FORM SASI-QA25
                                                  DEC 19, 1988

                             ANSYS QA NOTICE

                                                  NOTICE NO: QA95-04

SUBJECT:  Intel Pentium Floating Point Unit Flaw

DESCRIPTION:

As you may be aware, a hardware problem was discovered with the numeric
Floating Point Unit (FPU) of the earlier versions of the Intel Pentium
microprocessor.  The problem occurs when performing floating point
division on certain types of numbers.  We describe below relevant testing
carried out by ANSYS and Intel Corporation.  The goal of this testing was
to determine the possible impact of the flaw on your past and present
ANSYS work.  This QA Notice updates preliminary information communicated
by ANSYS, Inc. shortly after initial reports of the flaw were made
public.

ANSYS, Inc. ran our entire quality assurance acceptance test suite (1800
+ problems) on a machine containing a flawed Pentium (P5-90) chip using a
Rev. 5.1 version of the ANSYS program.  When the results were compared
with those from a 486-based computer, we saw no differences in the entire
set within a tolerance of 1x10-8, which is a stricter tolerance than we
normally use when comparing results from different machines.  This was
not a surprise since, based on Intel's white paper on the subject, the
likelihood of encountering a single divide flaw in randomly occurring
numerical divides is only one in 9 billion.  The number of divides that
occur in our entire ANSYS test suite is approximately 1.1 billion,
leaving a one in nine chance of randomly encountering one floating point
divide inaccuracy in the set.  Because this error is characterized by
probabilities of occurrence, we did an additional level of testing in
which the divide flaw could be intentionally "injected" during the
execution of the test.  This allowed us to study the impact of the flaw
on present and past ANSYS analyses.

    Intel's testing lab has the ability to intentionally inject divide
flaws into tests being performed.  We ran a suite of tests where we
intentionally injected 100 divide flaws per ANSYS run.  (it should be
noted that the flaw was injected in the worst possible place, the
twelveth bit.) This compares to an average of 670,000 floating point
divides and 48 million floating point instructions per test case (mostly
small to medium sized ANSYS runs).  As was expected from such an
experiment, we did see numerical differences between our baseline test
suite and those which had 100 divide inaccuracies within the run.  Even
with 100 flaws per test in our test set, we found that half of the tests
showed no differences within a tolerance of 0.01 % (1x10-4).  The
remainder of the tests showed imprecision of various levels in some
solution results and in some cases showed iterative solutions that did
not converge, especially for nonlinear analyses.  A small number of the

tests that were run with the intentionally injected divide flaw
terminated abnormally (because of nonconvergence) or took much longer to
achieve convergence.

To summarize our findings:

1.   Even with 100 divide inaccuracies artificially injected per test
          (an "impossible" situation to occur in real life, based on the
          probability), all results that we observed had accuracies that
          never dropped below two significant digits.

2.   We concur with Intel's initial finding that the likelihood of this
          flaw to occur is small.  Moreover the probable impact of such a
          flaw (one that has not been intentionally injected) on an ANSYS
          solution is even smaller.  Intel claims that, to date, only two
          users worldwide have had their work impacted by the divide
          flaw.

3.   We realize that this notice is not sufficient assurance for people
          doing safety-critical analysis.  We hope it provides some sense
          of the possible magnitude of any inaccuracies that this Pentium
          flaw could have caused in a worst-case situation.

4.   We recommend that you request a new chip from Intel.  The decision
     as to whether you need to re-perform analyses done on flawed Pentium
     chips lies with you, based on the criticality of any analyses you
     may have performed.

AFFECTED VERSIONS: N/A - Hardware Problem

AUTHOR:                                      DATE: April 10, 1995
                    Joe Manich

REVIEWED BY QA:                              DATE: April 10, 1995
                    Prashant Ambe

APPROVAL:                                    DATE: April 10, 1995
                    John A. Swanson

                                             SASI-QA3
Page 2 of 2                                  AUG. 29, 1993

*** END OF DOCUMENT ***

Page Last Reviewed/Updated Thursday, March 29, 2012