The program

An integrator,and very little else

GROMACS moves particles under a force field. Everything it is famous for follows from doing that quickly.

Classical, not quantum

Atoms are balls on springs. No electrons, which is why millions of particles are affordable.

Free and open

LGPL, with the source in the open. Nothing about a licence server ever enters the run.

Tuned to the metal

Hand-written SIMD kernels for the inner loop, which is most of where the speed comes from.

What it computes

Three answerspeople actually want

A trajectory is rarely the point. These are the questions the trajectory is there to answer.

How does it moveShort steps at femtosecond resolution, so bond vibrations and side-chain motion are resolved rather than averaged.
Where does it settleEnergy minimisation drops a built structure into a physically sane configuration before any dynamics start.
What does it costFree-energy methods — thermodynamic integration, BAR — turn a trajectory into a binding or solvation number.
Why it is fast

Four layersof parallelism

GROMACS is fast because it splits the same step four different ways, and each split is independent of the others.

SIMD inside a core

The non-bonded kernel is written against vector instructions, not left to the compiler to guess at.

Threads across cores

OpenMP splits the work inside one machine, which is the level most single-machine runs stop at.

MPI across nodes

Domain decomposition cuts the box into regions, one per rank, and only the boundaries are exchanged.

Offload to a GPU

Non-bonded forces and PME move to the card while the CPU keeps the bonded terms and the integration.

The practical consequence is that the same input runs on a laptop and on a rented GPU machine without changes. You are choosing how much hardware to point at it, not rewriting the run.

Honestly

What it willnot do for you

The engine is narrow on purpose, and the gaps are where most of the work actually goes.

NoBond breakingclassical force fields hold bonds fixed
ElsewhereTopology buildingpdb2gmx and CHARMM-GUI do that part
YoursKnowing when to stopconvergence is a judgement, not a flag

GROMACS will not choose your force field, build your topology, or tell you a run is too short. It integrates what you hand it, at the speed the hardware allows, and reports what happened. Preparation and analysis are separate jobs, and they are where results are usually won or lost.

On a machine

GROMACS isalready built

Nothing here needs compiling. Pick a machine, and these two commands put it on there.

hub://gromacs

$ dxflow workflow create --identity gromacs hub://gromacs

$ dxflow workflow start gromacs

Read it, then run itPrepaid and by the hour, on a machine that is yours about a minute after you ask.