runFEKO - ERROR 3623: FEKO caught an illegal instruction, code is not running on this processor

Altair Forum User
Altair Forum User
Altair Employee
edited October 2020 in Community Q&A

My processors are dual Xeon E5-2540 v3 (2 processors),I try to run FEKO at local machine.

In component launch options -> FEKO tag, if I set processers more than 1 (default is 16),  output tag which in window executing runfeko show:

 

Creating file startup_fr_1_ada_1.fek, please be patient ...

File startup_fr_1_ada_1.fek has been created:

58 metallic triangles in free space

FEKO caught signal 4 (PID 5152)

The following message from the master process (MYID= 0):

ERROR 3623: FEKO caught an illegal instruction, code is not running on this processor

 

The FEKO model that I used is “startup.fek”.

Selection 'Use shared memory / OpenMP .........' is useless.

 

Thanks

 

Tagged:

Answers

  • JIF
    JIF
    Altair Employee
    edited October 2018

    Hello Simba,

     

    Selection 'Use shared memory / OpenMP .........' is useless. 

    I would not recommend using that option. OpenMP is not used by many parts of the solver and for the last year (probably longer), using Intel MPI showed better performance, even on a single node. It was a very nice feature in the past that showed good performance gain when compared to MPI.

     

    What version of Feko are you using? Please provide more info such as the version and the .out file and full stdout that is produced.

     

    The error message is usually not associated with new CPUs, but old CPUs (AMD Athlon, Pentium II, etc.) that are not supported any more. If you have a ' Xeon E5-2540 v3' CPU, I don't expect this error. I have also not seen anyone else running into a similar problem.

  • Altair Forum User
    Altair Forum User
    Altair Employee
    edited October 2018

    Hello Simba,

     

    I would not recommend using that option. OpenMP is not used by many parts of the solver and for the last year (probably longer), using Intel MPI showed better performance, even on a single node. It was a very nice feature in the past that showed good performance gain when compared to MPI.

     

    What version of Feko are you using? Please provide more info such as the version and the .out file and full stdout that is produced.

     

    The error message is usually not associated with new CPUs, but old CPUs (AMD Athlon, Pentium II, etc.) that are not supported any more. If you have a ' Xeon E5-2540 v3' CPU, I don't expect this error. I have also not seen anyone else running into a similar problem.

    Hi JIF,

    Thanks for your suggestion.

    I Restore defaults 'Component launch options' , and run FEKO.

    Here are my FEKO Version(2018-2) and OS is Windows Server 2008. RAM = 64GB.

    <?xml version="1.0" encoding="UTF-8"?>FEKO_And_OS_Version.thumb.png.c47d5e360e154618d0e85aab7320e7f0.png

     

    FEKO_Component_Settins.png.1ca9293fad5859a487389fc6a4450367.png

    FEKO_Machines_Setting.png.a1faee841b180865fd4227b890dae694.png

     

    I didn`t change Other settings. I installed FEKO in today morning.

     

    And I use hotkey Alt+2 to run preFEKO, it sucessed.

     

    But Alt+4 runFEKO, ERROR 3623 is occurred.

    runFEKO.png.43a889d5f9acf2ccdc3408c1647f4496.png

     

    And here are my Model Files And Output Files in FDAM.zip

     

    Thanks you.

    Unable to find an attachment - read this blog

  • JIF
    JIF
    Altair Employee
    edited October 2018

    Hello Simba,

     

    I have asked some advice from the developers since we have not seen this problem from any other users.

    I'll keep you posted on progress.

  • Altair Forum User
    Altair Forum User
    Altair Employee
    edited October 2018

    Hello Simba,

     

    I have asked some advice from the developers since we have not seen this problem from any other users.

    I'll keep you posted on progress.

    Hi JIF,

     

    I keep paying attention to the forum. If you need more data or output files, please contact me.

  • JIF
    JIF
    Altair Employee
    edited October 2018

    I have sent you a private mail with some requests to try to track down the origin of the problem.

  • Altair Forum User
    Altair Forum User
    Altair Employee
    edited October 2018

    I have sent you a private mail with some requests to try to track down the origin of the problem.

    I upload my file by the link from your email.

  • JIF
    JIF
    Altair Employee
    edited October 2018

    Hello,

    The development team seems to think this is a bug in the Intel math library. For some reason, on some architectures, the condition number estimation seems to fail. It was previously found only on Linux parallel runs, but now we see that it sometimes also affects Windows. A fix that skips the condition number calculation will be available in the next update (should be out in about 2 weeks). I hope that you are able to update and wait for the fix.