Benchmark: OpenMM on Dihydrofolate Reductase (DHFR)


Benchmark Settings

OpenMM was benchmarked on the dihydrofolate reductase (DHFR) models taken from the Joint Amber/Charmm benchmark. This is a 159 residue protein with 2489 atoms. The version used for explicit solvent simulations included 7023 TIP3P water molecules, giving a total of 23,558 atoms. All simulations used the AMBER99SB force field and were run with OpenMM 7.6.

Treatment of Solvent

Three different methods were used to calculate the effect of solvent:

  1. Implicit: Solvent was represented with an OBC-GBSA implicit solvent model. Long range interactions were cut off at 2 nm.

  2. Explicit-RF: Solvent was modeled explicitly. Long range interactions were cut off at 1 nm using the reaction field method.

  3. Explicit-PME: Solvent was modeled explicitly. Long range interactions were calculated using Particle Mesh Ewald. The Ewald error tolerance was set to 0.0005, and the direct space cutoff was chosen to optimize performance (0.9 nm for the GPU simulations, 0.8 nm for the CPU simulations).

Integration

All simulations used LangevinMiddleIntegrator with a step size of 4 fs. The lengths of bonds involving hydrogen were constrained and the constraint error tolerance was set to 0.00001. Water molecules were fully rigid. Mass repartitioning was used to increase the mass of hydrogen atoms to 1.5 amu.

Platforms

Simulations were run using the following platforms and hardware:

  1. CUDA platform on an NVIDIA Titan V GPU.
  2. OpenCL platform on NVIDIA Titan V GPU.
  3. CPU platform on a 4 core 4.3 GHz Intel Core i7-7740X CPU.


Results

Performance is measured in ns/day.

CUDA Titan V

OpenCL Titan V

CPU

Implicit

1940

1137

21.1

Explicit-RF

1227

1034

39.2

Explicit-PME

925

590

34.2


BenchmarkOpenMMDHFR (last edited 2021-10-04 21:52:15 by peastman)