AMBER Archive (2009)

Subject: RE: [AMBER] -np not 2^x a problem w/ sander??? update

From: Ross Walker (ross_at_rosswalker.co.uk)
Date: Sat Nov 07 2009 - 10:31:36 CST


Hi Dean,

This means the MPI library you are using does not support MPI_IN_PLACE even
though it is MPI2 and part of the standard. Use -mpich instead of -mpich2 as
the argument to configure. Alternatively edit your config.h file and remove
-DUSE_MPI_IN_PLACE

Then make clean and recompile.

This should, I think, fix it.

All the best
Ross

> -----Original Message-----
> From: amber-bounces_at_ambermd.org [mailto:amber-bounces_at_ambermd.org] On
> Behalf Of Dean Cuebas
> Sent: Friday, November 06, 2009 11:47 AM
> To: AMBER Mailing List
> Subject: [AMBER] -np not 2^x a problem w/ sander??? update
>
> UPDATE:
>
> Actually, the job was setting up, but then stopped with this job error:
>
> 0 - MPI_REDUCE_SCATTER : Invalid buffer pointer: Arguments must specify
> different buffers (no aliasing)
> [0] [] Aborting Program!
> done.
>
> Any suggestions?
>
> Thanks!
>
> Dean
>
> __________________________________________________________
> Dear amber people,
>
> The place I'm running sander.MPI limits np to 48.
>
> I'm seeing a big performance hit with np 48 compared to np 32.
>
> I don't see this performance hit with pmemd. Is this normal?
>
> I'm doing SGLD, so I'm using sander.MPI
>
> Thanks very much!!!
>
> Dean
>
>
>
> _______________________________________________
> AMBER mailing list
> AMBER_at_ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber

_______________________________________________
AMBER mailing list
AMBER_at_ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber