Hi,
I'm not sure how much the Accelerated MD code has been tested ...
I noticed that the amd.log file indicates that vastly different things
are being done depending on whether using sander, pmemd, or
pmemd.cuda.
It looks like sander(.MPI) and pmemd.cuda are producing consistent,
reasonable results. The serial pmemd produces incorrect potential
energies in the amd.log file but seems to continue running. pmemd.MPI
also does not report correct potential energies in the amd.log file and it
crashes after a few thousand steps with a vlimit error.
Hopefully someone more familiar with the code can diagnose this (see below).
--Niel
My computer:
- Dual-Hex Intel Xeon Westmere X5660
- Tesla M2090 GPU
- Amber12 (bugfixed) compiled with Intel 12.1.3.293
- Mvapich2-1.7
Here is a snippet of the first 5 lines of the amd.log files for sander/pmemd (using identical input files):
----- sander -------
500 500 1 -25007.690267452886 97.756503548357 0.266050454349 0.034187216512 1169.152755500260 16.981277998957
500 500 2 -24690.817807373063 101.405382803730 0.332930689536 0.059337198428 797.638153040664 11.100091477062
500 500 3 -24267.077501445114 106.304598903008 0.470460860738 0.180232720594 369.977314168554 3.686330625345
500 500 4 -23833.546422042167 111.879880892760 0.724433898035 1.000000000000 66.969995258098 0.000000000000
500 500 5 -23529.988269129277 114.735166671938 1.000000000000 1.000000000000 0.000000000000 0.000000000000
----- pmemd -------
500 500 1 -4035.710352328125 97.739078899237 1.000000000000 0.034108678022 0.000000000000 17.009682999221
500 500 2 -4025.014279470225 101.728766328767 1.000000000000 0.062822293925 0.000000000000 10.587609924899
500 500 3 -4015.623017369073 106.687902362526 1.000000000000 0.203160613944 0.000000000000 3.163186263034
500 500 4 -4011.763895364519 111.525231242301 1.000000000000 1.000000000000 0.000000000000 0.000000000000
500 500 5 -4014.053067456131 112.584555024514 1.000000000000 1.000000000000 0.000000000000 0.000000000000
------ pmemd.MPI ------
500 500 1 -165212.893483722641 0.000000000000 0.000113479697 0.000615075147 236343.455471183988 181.217792148042
500 500 2 -165212.898910553078 0.000000000000 0.000113479689 0.000615075147 236343.464629531227 181.217792148042
500 500 3 -165212.903962433076 0.000000000000 0.000113479681 0.000615075147 236343.473155110085 181.217792148042
500 500 4 -165212.908289464307 0.000000000000 0.000113479674 0.000615075147 236343.480457430473 181.217792148042
500 500 5 -165212.911536836124 0.000000000000 0.000113479668 0.000615075147 236343.485937711986 181.217792148042
------ pmemd.cuda ------
500 500 1 -25006.887412592856 97.708975324000 0.266203374530 0.033973627781 1168.228358557432 17.060615439141
500 500 2 -24687.746908077854 101.242530671000 0.333745012483 0.057690924655 794.096550679039 11.360105169050
500 500 3 -24262.692940347857 105.747059168000 0.472521234810 0.153231057002 365.691248044969 4.471448308627
500 500 4 -23829.510644879862 110.686258329000 0.727710627560 1.000000000000 65.111177388485 0.000000000000
500 500 5 -23533.890157990856 113.611192760000 1.000000000000 1.000000000000 0.000000000000 0.000000000000
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Thu Jul 26 2012 - 13:30:02 PDT
