CHARMM Performance

Non-bonded interactions, and in particular solvent-solvent interactions, are the most computationally expensive part of standard classical molecular dynamics simulations. With current computer architectures it is cost-effective to use lookup tables for these interactions instead of calculating them each time, and the CHARMM implementation of non-bonded lookup tables (see J Comp Chem 30:1490 for implementation details, and more benchmarks) allows explicit solvent simulations to be run at least as fast with CHARMM as with GROMACS or NAMD. The following CHARMM input script, which also shows how to use the lookup tables in CHARMM versions 34-36, runs a 1000 step simulation of 28005 water molecules in a box at 0.12s/step on 12 Intel 2.4GHz cores (or at 0.14s/step on 8 cores with the PME setup):

*  28005 TIP3 BOX (96A)**3
*
read rtf  card name top_all22_prot.inp
read para card name par_all22_prot.inp
read sequence tip3 28005
gene wat noangle nodihe
read coor card name w28k-min.crd
crystal define ortho  96 96 96  90.0 90.0 90.0
crystal build cutoff 16.0
image byresidue sele resn tip3 end
shake bonh para
energy fshift vshift bycbim cutnb 16.0 cutim 16.0 ctofnb 12.0
! Alternatively, with PME:
! energy  eps 1.0 cutnb 13 cutim 13 ctofnb 9 ctonnb 7.5 vswi bycbim -
!     Ewald kappa 0.32 pmEwald order 4 fftx 90 ffty 90 fftz 90
lookup sele resn tip3 end
dynamics start nstep 1000 time 0.001 firstt 300.0 finalt 300.0 -
        inbfrq -1  imgfrq -1 echeck -1.0 ntrfrq 1000