[MITgcm-devel] SERIOUS performance problems with latest code

[mlosch at awi-bremerhaven.de]

For what it's worth:
On our XD1 with pgf77 and a 18CPU job, I don't observe any performance problems. However, I don't run the very latest 
code. My last entry in doc/tag-index is:

o multi-threading and IO:
  - replace some call to MDSIO S/R by the corresponding READ/WRITE_FLD S/R.
  - remove most of BEGIN_MASTER/ END_MASTER around READ/WRITE_FLD S/R calls
    (needed to work with SingleCpuIO): tested in all verif. exp.
o GAD: use simpler DST3 formulation as default (old version left within #ifdef)
  affects the results (but none of the verification test uses this scheme).
o Adding some pkg/diagnostics capability to pkg/seaice
  (will certainly be subject to changes)

checkpoint58q_post
...

I am rerunning a job with the very latest now ...
 

Martin Losch
Alfred Wegener Institute 
Postfach 120161, 27515 Bremerhaven, Germany; 
Tel./Fax: ++49(0471)4831-1872/1797



----- Original Message -----
From: Patrick Heimbach <heimbach at mit.edu>
Date: Tuesday, October 31, 2006 6:22 pm
Subject: Re: [MITgcm-devel] SERIOUS performance problems with latest code

> 
> Well,
> 
> > Hi D.,
> > Perhaps
> > >   God the code moves fast!
> > but the model doesn't!!
> 
> the code isn't God!
> 
> -p.
> 
> 
> 
> _______________________________________________
> MITgcm-devel mailing list
> MITgcm-devel at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-devel
>