[gmx-users] GROMACS Parallel Runs

Sunny ge_sunny at hotmail.com
Fri Oct 6 11:38:29 CEST 2006


Setting both fourier_nx and nkx to 256 and running grompp with -np 256 still 
produces the error:

could not find a grid spacing with nx and ny divisible by the number of 
nodes (256)

Does it mean 256 nodes are not allowed?

Sunny

>From: David van der Spoel <spoel at xray.bmc.uu.se>
>Reply-To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>Subject: Re: [gmx-users] GROMACS Parallel Runs
>Date: Fri, 06 Oct 2006 10:54:12 +0200
>
>Sunny wrote:
>>>From: "Dallas B. Warren" <Dallas.Warren at vcp.monash.edu.au>
>>>Reply-To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>>>To: Discussion list for GROMACS users <gmx-users at gromacs.org>
>>>Subject: RE: [gmx-users] GROMACS Parallel Runs
>>>Date: Fri, 06 Oct 2006 09:35:31 +1000
>>>
>>> > I have successfully run gmx on up to 128 cpus. When I scale
>>> > to 256 cpus, the
>>> > following error occurs. Does it mean that gmx can't be run on
>>> > 256 nodes?
>>> >
>>> > Fatal error:
>>> > could not find a grid spacing with nx and ny divisible by the
>>> > number of
>>> > nodes (256)
>>>
>>>Isn't that just due to the reasoning pointed out previously ....
>>>
>>
>>I think that's not the same problem. This time I have set fourier_nx=512 
>>(or nkx=512, I've tried both setting). When running grompp ... -np 256, it 
>>reports the error:
>>
>>Fatal error:
>>could not find a grid spacing with nx and ny divisible by the number of 
>>nodes (256)
>>
>>If I run the same grompp with the same .mdp file and less -np (i.e., 128), 
>>the error doesn't occur.
>>
>>Thanks,
>>
>>Sunny
>>
>>> >>the current version of gmx requires at least pme_order/2 grid points
>>> >>per
>>> >>processor
>>> >>for the x-dimension of the pme grid. With pme_order=4 and
>>>fourier_nx=64
>>> >>you end up with
>>> >>only one grid point per processor. On 64 CPUs you would have to set
>>> >>fourier_nx to 128
>>> >>then (or alternatively choose a smaller fourierspacing). These
>>>settings
>>> >>are not
>>> >>useful for 32 or less CPUs though, because they would unnecessarily
>>>slow
>>> >>down your
>>> >>calculation.

>set all of them to 256
>
>do not expect to get any performance gain though.
>
>--
>David.

_________________________________________________________________
All-in-one security and maintenance for your PC.  Get a free 90-day trial!  
http://clk.atdmt.com/MSN/go/msnnkwlo0050000001msn/direct/01/?href=http://www.windowsonecare.com/?sc_cid=msn_hotmail




More information about the gromacs.org_gmx-users mailing list