'%20d='M364.9%20143.2v98.9l-1.4%201.2-165.6%20140.4%2035.7%2029.7v-34.2l1.4-1.2%20123.4-105.4%206.5-5.6v37l-1.3%201.2-99.7%2085.8v84.7l-6.6-5.7-97-83.1-3.6-3%203.5-3%20174.4-150.6v-25.1L153.7%20360.5l-6.6%205.7v-98.9l1.4-1.2%20161.6-138-35.7-29.7v34.4L273%20134%20153.7%20236.7l-6.6%205.7v-37l1.3-1.2%2097-83.2V36.3L252%2042l97.1%2083.1%203.6%203-3.6%203-171.7%20147.7V304l106.2-90.6%2056.4-48.8%2018.3-15.7z'%20/%3e%3c/svg%3e)
Start from the system,not the benchmark
All of these integrate the same equations. What differs is which kind of system each was built around, and that is what should decide it.
Classical force fields, PME electrostatics, the same thermostats. Agreement between engines is expected.
One was built for proteins in water, another for metals, another for coarse-grained beads.
AMBER parameters run in GROMACS, CHARMM in several. The field is not locked to the engine.
Five that coveralmost everything
There are more, but a project that does not fit one of these is unusual enough that you already know why.
Four questions,in order
Ask these and the choice usually makes itself before any benchmark is run.
Protein in water, or a crystal lattice? That single fact removes most of the list immediately.
If a specific parameter set is required, use the engine it was published for and avoid the conversion.
Free energy, constant pH, QM/MM, enhanced sampling. Support for these is where engines genuinely diverge.
Working input files from a colleague beat any benchmark, especially on a first project.
The most underrated of these is the last one. Inheriting working input files and someone who can answer questions is worth more than a twenty per cent throughput difference, especially on a first project. Force fields are portable between engines far more often than habits are.
The tools youalso end up needing
No MD project is only an MD engine. These three categories account for most of the rest of the time.
Preparation — building topology, assigning parameters, solvating — is where mistakes get made and where the engines differ most in ergonomics. Analysis is MDAnalysis, MDTraj or the engine's own tools, and it is worth learning one properly rather than three badly. Visualisation is VMD, PyMOL or ChimeraX, and you will use it far more than you expect.
GROMACS isalready built
Nothing here needs compiling. Pick a machine, and these two commands put it on there.
$ dxflow workflow create --identity gromacs hub://gromacs
$ dxflow workflow start gromacs