[MITgcm-support] Equatorial Channel from -70 degree to +70 degree
mlosch at awi-bremerhaven.de
Wed Oct 4 02:42:18 EDT 2006
the results from
nPx=8, sNx=32, sNy=93 should be exactly the same (except for round-
off error) as those from
nPx=4, sNx=64, sNy=93,
nPx=2, sNx=128, sNy=93, and
nPx=1, sNx=256, sNy=93, regardless of time step.
When you have NaN's in one case and no NaN's in another, then
something else is wrong. What about your nSx/nSy? both should be 1 in
all of these examples so that sNx*nSx*nPx = 256, and sNy*nSy*nPy =
93 for all cases.
On 4 Oct 2006, at 08:29, Yuan Lian wrote:
> Hi, Dimitris,
> Thanks. I tried the configuration nPx=8, sNx=32, sNy=93 but the
> simulation gave NAN values. I had to reduce the number of grids in
> latitude. The configuration nPx=4, sNx=64, sNy=63 worked just fine.
> I will try nPx=4, sNx=64, sNy=93 and see how it works. I might try
> to use very short time step as well.
> Dimitris Menemenlis wrote:
>> Hi Yuan,
>>> I noticed that, in order to run the simulation, the number of
>>> grid points in
>>> X direction has to be greater than the number of grids in Y
>>> direction in a
>>> single tile.
>> There is no such restriction as far as possible model
>> configurations are concerned.
>>> The number of grids is 256x93
>>> So the question is: how to configure the "SIZE.h" to run the
>>> simulation with
>>> 4 or 8 processors?
>> Since 93 is not divisible by 2, and assuming you are using MPI,
>> you can have
>> nPx=4, sNx=64, sNy=93 and
>> nPx=8, sNx=32, sNy=93.
>> Cheers, Dimitris
>> MITgcm-support mailing list
>> MITgcm-support at mitgcm.org
> MITgcm-support mailing list
> MITgcm-support at mitgcm.org
More information about the MITgcm-support