[CIG-SEISMO] specfem3d_globe-7.0.0 overflow

[Seiji Tsuboi]

Dear Komatitsch-san

Thank you very much for your very quick reply to my inquiry.
Since I could not find where I should reduce DT in the program, I have
turned both USE_LDDRK and INCREASE_CFL_FOR_LDDRK on. Then computation
proceeds for NPROC=17 and NEX=816 case without overflow but the time per
iteration becomes almost 6 times longer. It may be necessary to compute for
more than one day. Is this what we expect? Should I turn USE_LDDRK false
and reduce DT instead?

Best regards,
Seiji Tsuboi

2018-07-03 6:02 GMT+09:00 Dimitri Komatitsch <komatitsch at lma.cnrs-mrs.fr>:

>
> Dear Tsuboi-san,
>
> It is probably the time step computed automatically by the code that is
> slightly above the CFL stability limit in that case. Thus, you probably
> just need to reduce the value of variable DT a bit (say 10% or so).
>
> Best regards,
> Dimitri.
>
> On 07/02/2018 09:41 AM, Seiji Tsuboi wrote:
>
>> Dear Sir,
>>
>> I have used specfem3d_globe-7.0.0 on intel cluster and have overflow
>> problems.
>> If I set NPROC=16 and NEX=768 (1536 cores), I could calculate seismograms
>> correctly. However if I increase NEX, for example NPROC=17, NEX=816 (1734
>> cores), xspecfem3D stops with overflow in the very beginning of
>> computation. ATTENUATION is set as true and it seems that the program
>> overflows in "multiply_accel_elastic" routine. I tried both ifort and
>> gfortran but the situation is the same. I would be grateful if someone
>> would let me know how I should solve this issue.
>>
>> Best regards,
>> Seiji Tsuboi
>>
>>
>> _______________________________________________
>> CIG-SEISMO mailing list
>> CIG-SEISMO at geodynamics.org
>> http://lists.geodynamics.org/cgi-bin/mailman/listinfo/cig-seismo
>>
>>
> --
> Dimitri Komatitsch, CNRS Research Director (DR CNRS)
> Laboratory of Mechanics and Acoustics, Marseille, France
> http://komatitsch.free.fr
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.geodynamics.org/pipermail/cig-seismo/attachments/20180703/57f3611f/attachment.html>