Folks -
Be aware, however, that at really high processor count with pmemd, using
nrespa > 1 tends to mess up the loadbalancing, and you will actually lose
ground on performance for particle mesh ewald (the exact value depends on
fft grid dimension maxima vs. processor count - nrespa stops working well
when the fft slab load is not distributed to all processors in the job);
since you also get slightly less accurate results from multiple timestep
protocols, you are then losing both performance and accuracy. For
generalized Born, I am sure nrespa is a performance help, regardless of
processor count. The next release of pmemd will probably do a bit better
for pme also when using nrespa > 1.
- Bob Duke
----- Original Message -----
From: "David A. Case" <case.scripps.edu>
To: <amber.scripps.edu>
Sent: Sunday, October 01, 2006 9:35 AM
Subject: Re: AMBER: Multiple Timestep MD
> On Wed, Sep 27, 2006, David LeBard wrote:
>>
>> I am interested in running a multiple timestep simulation, and I have
>> noticed that AMBER previously support such simulation techniques (using
>> ROAR). I wonder if it is possible to obtain an older copy of AMBER
>> (ver. 6 or 7) with this module included using my AMBER 8 license? If
>> not, is there still any possibility of obtaining a copy of the code and
>> a license of one of the older versions of AMBER?
>
> Both sander and pmemd (in the current versions of Amber) support multiple
> timestep techniques: see the "nrespa" variable.
>
> ....dac
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber.scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
>
-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Wed Oct 04 2006 - 06:07:05 PDT
