MOLECULAR DYNAMICS PERFORMANCE GUIDE - Digital Research Alliance of CANADA

BENCHMARK DETAILS

ID=240

  • Dataset: 6n4o
  • Software: NAMD3.cuda (binary_pack/3.0b3------)
  • Resource: 1 tasks, 4 cores, nodes, 4 GPUs, with NVLink
  • CPU: EPYC 7413 (Milan), 2.65 GHz
  • GPU: NVIDIA-A100-SXM4-40GB, 12 cores/GPU
  • Simulation speed: 76.9625 ns/day
  • Efficiency: 45.7 %
  • Site: Narval
  • Date: Sept. 7, 2023, 12:31 a.m.
  • Submission script:

    #!/bin/bash
    #SBATCH -c4 --gpus-per-node=a100:4
    #SBATCH --mem-per-cpu=2000 --time=1:0:0

    INPUT=namd3_input.in
    NAMDHOME=$HOME/NAMD_3.0b3_Linux-x86_64-multicore-CUDA

  • Notes:

    Performance 75.771000 ns/days, 4 GPUs, ng10503
    Performance 76.788900 ns/days, 4 GPUs, ng10503
    Performance 76.946200 ns/days, 4 GPUs, ng20503
    Performance 76.962500 ns/days, 4 GPUs, ng20503

  • Simulation input file:

    # Amber/(t,s,x)leap generated parm and crd file
    parmfile prmtop.parm7
    ambercoor inpcrd.rst7

    # Input
    bincoordinates equilibration.coor
    binvelocities equilibration.vel
    extendedsystem equilibration.xsc
    numsteps 100000
    outputtiming 100000

    # Output
    restartfreq 100000
    dcdfreq 100000
    outputEnergies 2000
    outputPressure 100000
    outputname _prod

    # AMBER FF settings
    amber on
    rigidBonds all
    useSettle on
    rigidTolerance 1.0e-8
    cutoff 8.0
    pairlistdist 11.0
    switching off
    exclude scaled1-4
    readexclusions yes
    1-4scaling 0.83333333
    scnb 2.0
    ljcorrection on

    # Integrator Parameters
    timestep 1.0
    nonbondedFreq 1
    fullElectFrequency 1
    stepspercycle 10
    wrapAll on
    CUDASOAintegrate on

    # Temperature control
    langevin on
    langevinTemp 300
    langevinDamping 1.0

    # Pressure control
    useGroupPressure yes
    LangevinPiston on
    LangevinPistonTarget 1.0
    LangevinPistonPeriod 200
    LangevinPistonDecay 100
    LangevinPistonTemp 300