Allegro GPU node benchmarks

This page is for all members to add their personal benchmarks of systems on the Allegro cluster for future reference a qualitative comparisons.

Add a section for the system at hand and be sure to include all relevant details: number of atoms (include in title), ensemble, constraint/cut-off settings, integrator, forcefield, software/libraries (+versions) etc.

GB3 in water (14028 atoms)

NPT-ensemble:

Integrator:

Forcefield:
  • Protein: CHARMM22*
  • Water: TIP3P

constraints:
  • H-bonds
  • rigid water

Single precision

Software:
  • OpenMM-7.0.1.dev-5e86c4f
  • cuda-7.0
  • Parmed-2.5.1.0
  • mdtraj-1.7.2
  • python-2.7.11 and python-3.5.1

Benchmark was run with two different cutoffs for the non-bonded interactions (ParticleMeshEwald)

Node: gpu054 (GeForce GTX 780): for 1nm cutoff: ~67ns/day For 9.5Å cutoff: ~80ns/day

Node gpu071 (GeForce GTX 980): for 1nm cutoff: ~86ns/day For 9.5Å cutoff: ~96ns/day

Synaptotagmin-I C2a domain in water (27,200 atoms)

NPT-ensemble:
  • Andersen Thermostat@300K; collision freq. ps^-1
  • MonteCarloBarostat 1bar 300K
  • Non-bonded cutoff 1 nm

Integrator:

Forcefield:
  • Protein: CHARMM27
  • Water: TIP3P

constraints:
  • H-bonds
  • rigid water

Single precision

Node: gpu073

Software:
  • OpenMM-7.0rc1
  • cuda-7.0
  • Parmed-2.5.1
  • mdtraj-1.7.2
  • python-3.5.1

Speed: 90 ns/day
Topic revision: r4 - 03 Jun 2016, SimonOlsson
 
  • Printable version of this topic (p) Printable version of this topic (p)