[MMTK] mmtk performances

Konrad Hinsen hinsen at cnrs-orleans.fr
Fri Mar 30 15:55:12 UTC 2007


On 24.03.2007, at 18:19, Davide Ricci wrote:

> I'm wondering if you have more recent informations about MMTK  
> performances against other codes like AMBER,CHARM, GROMACS for  
> example. The only information I've found with Google is an old post  
> (2000). At that time MMTK was supposed to be 20% slower than CHARM.  
> What's the situation for the last version ?

The only speed-critical code that has ever changed in MMTK is the  
normal mode calculation, where the very latest releases use newer and  
faster LAPACK routines than their predecessors. Energy evaluation,  
MD, and energy minimization have not changed significantly since the  
earliest releases.

Note that speed comparisons are a very tricky business. Beyond the  
usual difficulties (differences between machines and compilers),  
there is the additional problem in simulation codes that it is very  
difficult to make them perform exactly the same calculation. Many  
details of the algorithms are undocumented, and it is not always  
clear what the various options do. To give an example, CHARMM (at  
least in the version that I worked with ten years ago) by default  
updates its nonbonded list only every 10 steps. This obviously speeds  
up the computations, but I don't know if anyone ever studied the  
consequences for MD trajectories. I also wonder how many CHARMM users  
are aware of this feature and how to switch it off.

In MMTK, all parameters have default values that are chosen  
conservatively for accuracy rather than for speed. In some cases  
(e.g. Ewald summation), the default is so conservative that  
performance is extremely bad, unless you fine-tune the parameters for  
your system.

Konrad.
--
---------------------------------------------------------------------
Konrad Hinsen
Centre de Biophysique Moléculaire, CNRS Orléans
Synchrotron Soleil - Division Expériences
Saint Aubin - BP 48
91192 Gif sur Yvette Cedex, France
Tel. +33-1 69 35 97 15
E-Mail: hinsen at cnrs-orleans.fr
---------------------------------------------------------------------



--
---------------------------------------------------------------------
Konrad Hinsen
Centre de Biophysique Moléculaire, CNRS Orléans
Synchrotron Soleil - Division Expériences
Saint Aubin - BP 48
91192 Gif sur Yvette Cedex, France
Tel. +33-1 69 35 97 15
E-Mail: hinsen at cnrs-orleans.fr
---------------------------------------------------------------------






More information about the mmtk mailing list