Re: [AMBER] AMBER qualitatively compared to other packages

From: Thomas Evangelidis <>
Date: Wed, 28 Mar 2018 13:38:44 +0200

On 27 March 2018 at 15:53, David A Case <> wrote:

> On Fri, Mar 23, 2018, Dmitry Suplatov wrote:
> >
> > I have been recently wondering about benefits of the AMBER program
> compared
> > to other software solutions (e.g., NAMD, Gromacs).
> Your email arrived at a very busy time (we are preparing for a big
> release), and has no simple answers. I'll try to make a few comments
> below.
> > 1. My understanding is that AMBER is not the fastest implementation,
> i.e.,
> > the GROMAC's GPU version of classical MD is significantly faster, right?
> Sigh...comparing implementation speeds could be a full-time job. One
> thing to remember is that Amber's GPU code (pmemd.cuda) does almost all
> of the calculation on the GPU, so that performance is almost entirely a
> function of which GPU one is using, and does not depend on having a
> high-performance (or expensive) CPU. There are lots of benchmarks at
> the GPU page of I hope these may be of some help.
​GROMACS and NAMD are no match for AMBER (pmemd.cuda) in terms of speed on
GPUs! OpenMM has comparable performance on all-atom simulations
but slows down in implicit solvent simulation when you increase the cutoff
for the electrostatics (pmemd.cuda works with infinite cutoff in implicit
solvent). To sum up, AMBER is the fastest classical MD engine I am aware of
on GPUs. The only limitation is that not all the algorithms that are
available in sander are also available in pmemd.cuda, but this difference
tends to be balanced as new software releases are coming out.

> >
> > 2. What about the implementation of the force fields?
> > - Both GROMACS and NAMD support the AMBERFFs, is there any difference in
> > how they do it compared to the original AMBER?
​A few years ago I was trying to reproduce energies in NAMD using AMBER
force fields and implicit solvent but they were totally uncorrelated. The
same happened with MM/GBSA free energies of binding of ligands. Those that
I obtained using NAMD trajectories were rubbish. Since then I don't trust
implementations of AMBER ff in other MD engines, but this is rather a
personal choice.

Dr Thomas Evangelidis
Post-doctoral Researcher
CEITEC - Central European Institute of Technology
Masaryk University
Kamenice 5/A35/2S049,
62500 Brno, Czech Republic

AMBER mailing list
Received on Wed Mar 28 2018 - 05:00:03 PDT
Custom Search