Hi Ross,
Thanx for the reply. Perhaps I could rephrase. My QM system contains around 53 atoms and my MM region around 9500 (including explicit solvent). The region of interest is a divalent metal coordinated to 4 waters and an NTP. The parameters for the NTP are taken from Bryces webpage and I believe they are well developed and commonly used within amber. Now, I wonder would it be scientifically valid to model the divalent metal and 4 coordinating waters using QM and model the NTP with the classical MM, using Bryces parameters? I'd imagine that reducing the QM region by around 40 atoms would surely be beneficial for performance in sander, but I wonder will the QM region take into account the MM residues and NTP and indeed, vice versa.
Looking forward to your (and other) replies.
Maura
________________________________________
From: Ross Walker [ross.rosswalker.co.uk]
Sent: 12 August 2011 00:18
To: 'AMBER Mailing List'
Subject: Re: [AMBER] QM runs
> Is it true that one cannot perform qmmm calculations with the pmemd
> module, even in the current Amber11? If so, why is this? Currently
Yes it is true.
Why? - PMEMD is designed to run very fast and in a highly parallel fashion.
Unfortunately the QM code is nowhere near as parallel - because it involves
things like matrix diagonalizations etc. Adding it to PMEMD would thus
destroy most of the advantages (given the QM part is generally the slowest
step anyway) and so it has not been done. Unfortunately this is what happens
when machines become more and more parallel. It becomes harder and harder to
preserve performance.
We could spend a significant amount of time improving the QM/MM support and
finding a way to include it in PMEMD in a way that does not hurt the
classical performance etc but unfortunately man power is very finite these
days.
> I'm using sander, but after using pmemd for classical simulations, I
> notice the huge decrease in performance. Is sander the only option
> here for qmmm MD jobs or can pmemd be implemented? Is there any advice
> on how to speed up the qm run?
What does your QM run look like? How have you set it up, how many QM atoms?
MM atoms? - did you compile with Intel MKL?
There are a VERY large number of degrees of freedom here with regards to
performance etc.
One thing to try is to compile, linking against MKL and then set
diag_routine=0 which will then test the various different diagonalizers in
the MKL Lapack and choose the fastest one.
All the best
Ross
/\
\/
|\oss Walker
---------------------------------------------------------
| Assistant Research Professor |
| San Diego Supercomputer Center |
| Adjunct Assistant Professor |
| Dept. of Chemistry and Biochemistry |
| University of California San Diego |
| NVIDIA Fellow |
|
http://www.rosswalker.co.uk |
http://www.wmd-lab.org/ |
| Tel: +1 858 822 0854 | EMail:- ross.rosswalker.co.uk |
---------------------------------------------------------
Note: Electronic Mail is not secure, has no guarantee of delivery, may not
be read every day, and should not be used for urgent or sensitive issues.
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Aug 12 2011 - 04:00:03 PDT
