[MITgcm-devel] cs510 on IBM p690

[Martin Losch]

Hi there,

I am having difficulties running the high resolution cs510  
experiment, that Dimitris so nicely prepared and documented on  
MITgcm_contrib (no irony there!)

I start from rest and after switching off the seaice model, or more  
specifically seaice dynamics, also after switching to the EVP solver  
I finally managed to get 1 full time step without the model blowing  
up completely so that I at least get some output. I have attached Eta  
after one timestep. I use SIZE.h_27 (216 tile on 27 cpus), see  
attached code directory and slightly modified *_OPTIONS.h files  
(mainly SEAICE). I have turned off all PTRACERS/GCHEM stuff (even in  
packages.conf, although not in the attached code.tgz).

Clearly something is wrong with the exchanges. I cannot reproduce  
this result with a cs32 configuration: It runs fine without any  
obvious problems.

For this I use checkpoint58v_post code.

I can reproduce this on two different IBM p690 ( http://jumpdoc.fz- 
juelich.de/  the FZJ link on that page and http://www.hlrn.de/doc/ 
basics.html#id2449391 ) with xlf95.

Any idea, what might be going wrong, what I could try next?
Anyone with experience on thes p690?

Despairing,
Martin

PS. and as usual, I will not be able to provide access to any of  
these machines (o:

-------------- next part --------------
A non-text attachment was scrubbed...
Name: code.tgz
Type: application/octet-stream
Size: 59981 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-devel/attachments/20070213/01a6a771/attachment.obj>
-------------- next part --------------

-------------- next part --------------
A non-text attachment was scrubbed...
Name: eta.png
Type: image/png
Size: 143200 bytes
Desc: not available
URL: <http://mitgcm.org/pipermail/mitgcm-devel/attachments/20070213/01a6a771/attachment.png>