I set-up a system using hydrogen mass repartitioning (HMR) using a 4fs
timestep and ran using 24 processors for 4 ns simulation time (total time =
11.5 hours). When comparing it against the 2fs simulation it took longer
(total time = 10.6 hours). A setting might be wrong somewhere, but it is
not obvious.
Main changes from 4 fs sims from 2 fs sims:
1. Identical box size and number of waters added.
original 2fs Box size =
Total vdw box size: 67.818 91.065 92.217 angstroms.
Volume: 569515.132 A^3
new 4fs Box size:
Total vdw box size: 67.818 91.065 92.217 angstroms.
Volume: 569515.132 A^3
2. Use of new montecarlo barostat
3. changing dt from 0.002 to 0.004
4. change cut from 8.0 to 9.0
5. setting ntave to 0
6. some changes to print such as NTPR and NTWR
>From the timings, I think this might be an issue with the mpi configuration
as the average ns/day seems to indicate a speed up, but the wallclock tells
a different story. However, I would like to make sure that my settings are
correct. Particularly for ntave and mcbarint. If anyone has any ideas, that
would be great
######################
#HMR input file:
&cntrl
nstlim=2000000, dt=0.004, ntx=5, irest=1,
ntpr=1000, ntwr=2000000, ntwx=500,
temp0=300.0, ntt=1, tautp=2.0, ioutfm=1, ntxo=2, ig=-1, cut=9.0,
barostat=2, mcbarint=100, ntave=0,
ntb=2, ntp=1,
ntc=2, ntf=2,
nrespa=1,
&end
######################
#
#
######################
#normal 2 fs input file:
&cntrl
nstlim=2000000, dt=0.002, ntx=5, irest=1,
ntpr=500, ntwr=1000000, ntwx=500,
temp0=300.0, ntt=1, tautp=2.0, ioutfm=1, ntxo=2, ig=-1,
ntb=2, ntp=1,
ntc=2, ntf=2,
nrespa=1,
&end
######################
######################
######################
######################
Timings for 2 fs:
| NonSetup CPU Time in Major Routines, Average for All Tasks:
|
| Routine Sec %
| ------------------------------
| DataDistrib 6023.97 15.84
| Nonbond 30754.53 80.88
| Bond 6.45 0.02
| Angle 73.44 0.19
| Dihedral 250.79 0.66
| Shake 199.60 0.52
| RunMD 712.39 1.87
| Other 2.04 0.01
| ------------------------------
| Total 38023.21
| PME Nonbond Pairlist CPU Time, Average for All Tasks:
|
| Routine Sec %
| ---------------------------------
| Set Up Cit 344.21 0.91
| Build List 3158.82 8.31
| ---------------------------------
| Total 3503.03 9.21
| PME Direct Force CPU Time, Average for All Tasks:
|
| Routine Sec %
| ---------------------------------
| NonBonded Calc 21337.98 56.12
| Exclude Masked 458.01 1.20
| Other 1239.56 3.26
| ---------------------------------
| Total 23035.55 60.58
| PME Load Balancing CPU Time, Average for All Tasks:
|
| Routine Sec %
| ------------------------------------
| Atom Reassign 2.96 0.01
| Image Reassign 6.25 0.02
| FFT Reassign 0.01 0.00
| ------------------------------------
| Total 9.22 0.02
| Final Performance Info:
| -----------------------------------------------------
| Average timings for last 2000 steps:
| Elapsed(s) = 36.86 Per Step(ms) = 18.43
| ns/day = 9.38 seconds/ns = 9214.24
|
| Average timings for all steps:
| Elapsed(s) = 38042.45 Per Step(ms) = 19.02
| ns/day = 9.08 seconds/ns = 9510.61
| -----------------------------------------------------
| Master Setup CPU time: 4.10 seconds
| Master NonSetup CPU time: 37949.66 seconds
| Master Total CPU time: 37953.76 seconds 10.54 hours
| Master Setup wall time: 6 seconds
| Master NonSetup wall time: 38043 seconds
| Master Total wall time: 38049 seconds 10.57 hours
#####################################################################
#####################################################################
#####################################################################
#####################################################################
Timings for 4 fs:
| NonSetup CPU Time in Major Routines, Average for All Tasks:
|
| Routine Sec %
| ------------------------------
| DataDistrib 6016.64 14.50
| Nonbond 34083.64 82.12
| Bond 6.95 0.02
| Angle 75.69 0.18
| Dihedral 257.11 0.62
| Shake 209.68 0.51
| RunMD 732.79 1.77
| Other 122.96 0.30
| ------------------------------
| Total 41505.46
| PME Nonbond Pairlist CPU Time, Average for All Tasks:
|
| Routine Sec %
| ---------------------------------
| Set Up Cit 731.30 1.76
| Build List 7938.13 19.13
| ---------------------------------
| Total 8669.43 20.89
| PME Direct Force CPU Time, Average for All Tasks:
|
| Routine Sec %
| ---------------------------------
| NonBonded Calc 20390.62 49.13
| Exclude Masked 523.13 1.26
| Other 1535.61 3.70
| ---------------------------------
| Total 22449.36 54.09
| PME Reciprocal Force CPU Time, Average for All Tasks:
|
| Routine Sec %
| ---------------------------------
| 1D bspline 319.85 0.77
| Grid Charges 410.92 0.99
| Scalar Sum 93.01 0.22
| Gradient Sum 651.85 1.57
| FFT 1459.25 3.52
| ---------------------------------
| Total 2934.88 7.07
| PME Load Balancing CPU Time, Average for All Tasks:
|
| Routine Sec %
| ------------------------------------
| Atom Reassign 8.20 0.02
| Image Reassign 15.61 0.04
| FFT Reassign 0.01 0.00
| ------------------------------------
| Total 23.82 0.06
| Final Performance Info:
| -----------------------------------------------------
| Average timings for last 1000 steps:
| Elapsed(s) = 20.75 Per Step(ms) = 20.75
| ns/day = 16.66 seconds/ns = 5187.32
|
| Average timings for all steps:
| Elapsed(s) = 41536.36 Per Step(ms) = 20.77
| ns/day = 16.64 seconds/ns = 5192.05
| -----------------------------------------------------
| Master Setup CPU time: 6.76 seconds
| Master NonSetup CPU time: 41457.54 seconds
| Master Total CPU time: 41464.30 seconds 11.52 hours
| Master Setup wall time: 8 seconds
| Master NonSetup wall time: 41537 seconds
| Master Total wall time: 41545 seconds 11.54 hours
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Wed Aug 05 2015 - 16:30:03 PDT
