Benchmark

We show benchmarks of GENESIS (versions 1.0 and 1.1.0) on Intel with GPU machines and K computer.

Condition

Integrator : VVER or RESPA (PME every 2 steps)
Ensemble : NVE
Time step : 2 fs

  • Intel and GPU benchmarks (CPU, CPU+GPU)
    CPU: Intel(R) Xeon(R) CPU E5-2670 0 @ 2.60GHz (2 CPU x 8 cores/node)
    GPU: NVIDIA Tesla K20x (2 GPUs/node)
    Compiler: openmpi 1.10.1 + intel (fortran/C) 15.0.5.223 + cuda 6.5
    Precision : double precision (dp) or single precision (sp)
  • K computer benchmarks
    CPU: SPARC64(TM) VIIIFx (8 cores/node)
    Compiler: Fujitsu compiler
    Precision: double precision

Intel and GPU benchmarks (CPU, CPU+GPU)

  • MSPA (216,726 atoms) uun_NVE
    CHARMM c36FF
    PME grid = 144,144,144
    switch, cutoff, pairlist= 10, 12, 13.5 Å

 

 

 

  • STMV (1,066,628 atoms) stmv_NVE
    CHARMM FF
    PME grid = 256, 256, 256
    switch, cutoff, pairlist = 9, 10, 11.5 Å

K computer benchmarks

  • 27 x STMV (28,798,956 atoms) stmv27_K_NVE
    CHARMM FF
    PME grid = 512, 512, 512
    switch, cutoff, pairlist = 9, 10, 11.5 Å