[CIG-SEISMO] Fwd: Running SpecFem Globe at 10seconds ?
komatitsch at lma.cnrs-mrs.fr
Tue May 12 06:08:59 PDT 2015
-------- Forwarded Message --------
Subject: Running SpecFem Globe at 10seconds ?
Date: Tue, 12 May 2015 15:01:50 +0200
From: masson <masson at ipgp.fr>
To: Dimitri Komatitsch <komatitsch at lma.cnrs-mrs.fr>
Is there any special things to do to scale up SpecFem Globe ? I used
the latest version (Git) with default setup (I changed nchunks to 6 and
pml to .false.). That worked fine. Then I used nex_xi=nex_eta=528 and
nproc_xi=n_proc_eta=22 compiled using intel on a cray system (hopper) at
NERSC and I am getting a "negative mass matrix term for the ocean". I
had similar problems with another earlier version I customized to
include the latest Berkeley 3D model (which we would be happy to share
once the model is published) it works fine using 300 total processors
but crashes at 2904. Did I miss something obvious here ?
Many thanks for helping,
Le 11.05.2015 15:26, Dimitri Komatitsch a écrit :
> Hi Yder, Hi all,
> Thanks! We fixed that in the devel branch a couple of weeks ago, I
> have just copied that change to the master branch thus if you type
> "git pull" you should be all set (otherwise please email me and I
> have a look).
> With the new robot (BuildBot) in place this should not happen any
> more in the future, we now have a very safe way of thoroughly
> new contributions (thanks to David Luet, who developed that).
> On 05/11/2015 12:29 PM, masson wrote:
>> Hi all,
>> I can't seem to compile the latest git for Specfem Globe. It looks
>> the variable 'I_should_read_the_database' ins't in the right _par
>> or the wrong module is loaded in src/specfem3D/get_attenuation.f90.
>> CIG-SEISMO mailing list
>> CIG-SEISMO at geodynamics.org
More information about the CIG-SEISMO