[MITgcm-support] How to calculate the RAM requirement?

[Mark Hadfield]

Göran Broström wrote:
> Hi all,  
>  
> To contribute to the confusion :) A long time ago I got a formula (from Chris) 
> size=nx*ny*nz*8*50 
> (grid points * 8 bits * 50 fields) 
> That would make about 10 GB. Perhaps it should be 100 fields in the new model 
> and with ice as suggested by Dimitris.
Do you mean 8 bytes?

A few months ago I measured the working set size of a simple 3D 
hydrodynamics-only run on 1 CPU and compared it with the size of a 3D 
double-precision array (nx*ny*nz*8 bytes). The ratio (let's call it R) 
varied between 34 and 71. The main determining factors were:

    * The timeave package is quite expensive, as you'd expect,
      increasing R by about 20.
    * Using a 2 subgrids in each direction (nSx = nSy = 2) reduces R by
      about 8-10 relative to a single-subgrid run. But this trend
      reverses when nSx or nSy is increased beyond about 2.

-- 
Mark Hadfield          "Ka puwaha te tai nei, Hoea tahi tatou"
m.hadfield at niwa.co.nz
National Institute for Water and Atmospheric Research (NIWA)


-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://mitgcm.org/pipermail/mitgcm-support/attachments/20070516/f7c5c617/attachment.htm>