[MITgcm-support] convincing compiler of node memory size

Matthew Mazloff mmazloff at ucsd.edu
Thu Apr 15 13:33:40 EDT 2010


Hi Santha,

Level 1 is stored in local cpu/node memory, while Level 2+ are written  
to external disk.  You can optimize your run efficiency by choosing a  
level 1 size that uses all available cpu memory.  (This is what Holly  
is trying to do.)

-Matt



On Apr 15, 2010, at 8:26 AM, Santha Akella wrote:

> I am able to build and run my adjoint model okay.
> Here is what I have in my tamc.h for the checkpointing.
> #ifdef ALLOW_TAMC_CHECKPOINTING
>
>       integer    nchklev_1
>       parameter( nchklev_1      =    16 )
>       integer    nchklev_2
>       parameter( nchklev_2      =    12 )
>       integer    nchklev_3
>       parameter( nchklev_3      =    10 )
>       integer    nchklev_4
>       parameter( nchklev_4      =    8 )
> ----
> as you see, i use other levels as well.
> Also I use only 120 cpus in my SIZE.h , and build with  
> linux_ia64_ifort+mpi_altix_nas
> Johnny Chang  @ NAS user services was very helpful. Hope that helps,
> Santha
>
>
>
> On Thu, Apr 15, 2010 at 11:14 AM, Martin Losch <Martin.Losch at awi.de>  
> wrote:
> Hi Holly,
>
> usually the default "memory model" only allows access to 2GB of RAM.  
> Look for a flag that lets you change the memory model to something  
> like "medium" (for pgf compilers I once used -mc_mode=medium, if I  
> remember correctly).
>
> Martin
>
> On Apr 15, 2010, at 5:00 PM, Holly Dail wrote:
>
> > Hello all -
> >
> > I'm trying to build an adjoint code for the Nehalem nodes on  
> Pleiades, and am getting the same error reported in January by  
> Santha here:
> > http://mitgcm.org/pipermail/mitgcm-support/2010-January/006434.html
> >
> > Reducing my nchklev_1 does allow a successful compile, but I don't  
> seem to be able to generate a binary that uses anywhere close to the  
> 3 GB available on the Nehalem nodes.  In my config, if I use  
> nchklev_1 = 30 I get a binary that uses nearly 1 GB, but if I set  
> nchklev_1 = 80 the link fails.
> >
> > The head nodes have 4 GB of memory, but perhaps there is a setting  
> somewhere telling the compiler 1 GB is the maximum?  I imagine I  
> could submit the compile to the queue, but is there another solution?
> >
> > I'm using linux_ia64_ifort+mpi_ice_nas with -xSSE4.2.
> >
> > Thanks -
> > Holly
> >
> > _______________________________________________
> > MITgcm-support mailing list
> > MITgcm-support at mitgcm.org
> > http://mitgcm.org/mailman/listinfo/mitgcm-support
>
>
> _______________________________________________
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
> http://mitgcm.org/mailman/listinfo/mitgcm-support
>
> <ATT00001..txt>

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20100415/87cf763f/attachment.htm>


More information about the MITgcm-support mailing list