[gmx-developers] MPI slower than Non-MPI gromacs

[Erik Lindahl]

Hi William,

The Gromacs demo is an extremely small system. Since the innerloops are 
coded in assembly it probably needs in the order of a millisecond to do 
a step, and your MPI latencies are likely higher than this (or at least 
in the same ballpark) if you are using the TCP/IP stack in windows and 
ethernet...

Test it with the DPPC membrane system from the benchmark pages at 
www.gromacs.org.

Cheers,

Erik

On Feb 18, 2004, at 3:26 AM, William Bagdan wrote:

> Hi
>
>  
>
> For win32 I have compiled MPICH, FFTW 2.1.5 with MPICH and then 
> Gromacs 3.1.4 with MPI and the FFTW with MPI.  After Testing the MPICH 
> and FFTW with MPI, I then began testing the GMXDemo using MPI.  The 
> results are not reassuring.  With MPI the time is longer than just 
> running straight Gromacs without the MPI.  Anybody have any ideas?  I 
> defined USE_MPI, but am not sure if I need to set some other defines 
> to get the FFTW.
>
>  
>
> If anybody is interested I would be willing to give you the setup for 
> compiling everything and the instructions.
>
>  
>
> William
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: text/enriched
Size: 1884 bytes
Desc: not available
URL: <http://maillist.sys.kth.se/pipermail/gromacs.org_gmx-developers/attachments/20040218/9894b40d/attachment.bin>