Performance#

Speed benchmarks#

The simulation speed of ACEMD was benchmarked against several systems at typical production conditions on different GPU devices. The GPU devices were operating under sustained load. The speed is measured in ns/day.

Device

Force field

System

DHFR

FactorIX

STMV

NVIDIA GeForce GTX 1080

AMBER

621

136

7.01

CHARMM

555

134

8.37

NVIDIA GeForce GTX 1080 Ti

AMBER

762

202

9.89

CHARMM

672

190

11.4

NVIDIA GeForce RTX 2080 Ti

AMBER

1040

313

15.4

CHARMM

979

296

17.0

NVIDIA GeForce RTX 3090

AMBER

1308

434

22.4

CHARMM

1258

416

25.1

NVIDIA GeForce RTX 4090

AMBER

2144

822

58.2

CHARMM

2051

791

56.8

The input files can be found in $CONDA_PREFIX/share/acemd after installing ACEMD.

Simulation systems:
Simulation conditions:
  • Force field:
  • Electrostatic interactions:
    • Particle-mesh Ewald (PME)

    • Grid spacing: <1.0 Å

    • Real-space cutoff: 9.0 Å

  • van der Walls interactions:
    • Cutoff: 9.0 Å

    • Switch function: none for AMBER or 7.5 Å for CHARMM

  • Constraints:
    • Constrained hydrogen bonds

    • Rigid water

    • Constraint tolerance: 1.0e-6

  • Integrator and thermostat:
    • Multi-step integrator period: 2

    • Time step: 4.0 fs

    • Temperature: 298.15 K

    • Friction: 0.1 ps^-1

    • Hydrogen mass: 4.0 a.u

  • Output:
    • Trajectory period: 100 ps (25000 steps)