[gmx-users] gromacs preformance versus amber

Mark Abraham Mark.Abraham at anu.edu.au
Wed Nov 21 22:03:06 CET 2007


servaas michielssens wrote:
> I am considering to use gromacs for MD-simulation for speed reasons, so
> I tried to compare a amber simulation to a gromacs simulation. I ran a
> 20ps simulation with amber. I converted my topology and coordinate file
> from amber to gromacs using and tried the same run in gromacs.
> amb2gmx.pl(http://chemistry.csulb.edu/ffamber/). This is my amber input:
> 
> 
> &cntrl
>   imin=0,nstlim=10000,nmropt=0
>   irest=1,ntx=5,
>   iwrap=1,
>   nscm=1000,
>   dt=0.002,
>   ntb=2,ntp=1,
>   ntt=1,tautp=1.0,temp0=300.0,
>   ntc=2,ntf=2,
>   cut=8.0,
>   ntwx=250,ntpr=250,
> /

It's been a while since I used AMBER, but this looks like straight 
cut-off electrostatics at 8 angstrom (vomit!). Thus you're comparing 
apples with oranges, since GROMACS is using PME. You will need a 
thorough understanding of the options in both AMBER and GROMACS to come 
up with a sensible simulation that is fairly comparable.

> This is my gromacs input:
> 
> title           = FWS
> cpp             = /usr/bin/cpp
> constraints     = all-bonds
> integrator      = md
> dt              = 0.002 ; ps !
> nsteps          = 10000 ; total 20 ps.
> nstcomm         = 1000
> nstxout         = 250 ; collect data every 1 ps
> nstlog          = 0
> nstenergy       = 250
> nstvout         = 0
> nstfout         = 0
> nstlist         = 10
> ns_type         = grid
> rlist           = 1.0
> coulombtype     = PME
> rcoulomb        = 1.0
> vdwtype         = cutoff
> rvdw            = 1.4
> fourierspacing  = 0.12
> fourier_nx      = 0
> fourier_ny      = 0
> fourier_nz      = 0
> pme_order       = 4
> ewald_rtol      = 1e-5
> optimize_fft    = yes
> ; Berendsen temperature coupling is on
> Tcoupl          = berendsen
> tau_t           = 1.0 1.0
> tcgrps          = protein non-protein
> ref_t           = 300 300
> ; Pressure coupling is on
> Pcoupl          = parrinellorahman
> tau_p           = 0.5
> compressibility = 4.5e-5
> ref_p           = 1.0
> ; Generate velocites is on at 300 K.
> gen_vel         = no
> gen_temp        = 300.0
> gen_seed        = 937475
>                                                                                                                                           
> 
> I ran it on 4 AMD Opteron(tm) Processor 250. It turned out that amber
> could simulate 1.5ns/day and gromacs only 1ns/day. This was a surprise
> to me, I expected that gromacs would be faster.  Could it be that using
> gromacs with the amberFF is not faster than just using amber? I am new
> to gromacs and perhaps I didn't compare them correct.

The rate-determining step in all MD simulations is the calculation of 
the non-bonded interactions. Once you choose comparable algorithms here, 
the assembly-optimized routines for GROMACS will likely stomp all over 
AMBER. The "implementation of the force field" is way down the list of 
importance.

You will also likely want to compare single-processor jobs.

Mark



More information about the gromacs.org_gmx-users mailing list