Hi,
thanks a lot for your answer, but I have some doubts
yet...
My .gro files are very different and I'm not sure that the
architecture (CPU / GPU) could influence the result so
much as in my example (the first is the output of the
serial code and the second that of the parallel ones).
SERIAL CODE:
1LYSH N 1 3.240 2.123 0.830 0.0828 0.7601
-0.2788
1LYSH CA 2 3.114 2.170 0.777 -0.1210 0.1179
-0.3660
1LYSH CB 3 3.060 2.258 0.889 -0.1060 0.0142
-0.2772
1LYSH CG 4 3.103 2.403 0.867 -1.1970 0.3922
0.0170
1LYSH CD 5 3.078 2.498 0.983 -0.5172 1.0015
-0.3267
1LYSH CE 6 3.120 2.451 1.122 -0.5706 0.5624
-0.4567
1LYSH NZ 7 3.024 2.505 1.220 0.2425 1.4466
-0.1367
1LYSH C 8 3.006 2.065 0.754 0.1697 -0.4396
0.7191
1LYSH O 9 2.896 2.086 0.701 0.0216 -1.3952
0.6269
2VAL N 10 3.036 1.943 0.799 0.5429 -0.4260
0.5099
2VAL H 11 3.132 1.933 0.831 0.8289 0.8067
0.0933
2VAL CA 12 2.955 1.826 0.822 0.2219 -0.3946
-0.4277
PARALLEL CODE:
1LYSH N 1 3.333 1.969 0.702 -0.4164 -0.1749
-0.1669
1LYSH CA 2 3.221 2.054 0.737 -0.0727 0.2241
-0.0385
1LYSH CB 3 3.163 2.122 0.613 -0.1387 -0.2965
-0.2935
1LYSH CG 4 3.219 2.261 0.586 0.0989 -0.2444
0.4520
1LYSH CD 5 3.100 2.352 0.558 0.5215 0.0874
-0.3041
1LYSH CE 6 3.123 2.486 0.627 -0.1705 0.3629
-0.6025
1LYSH NZ 7 2.996 2.514 0.696 -0.2146 0.6887
-0.8084
1LYSH C 8 3.110 1.973 0.800 -0.1975 -0.0906
-0.6520
1LYSH O 9 3.007 2.030 0.837 0.2640 0.4243
-0.1313
2VAL N 10 3.141 1.849 0.840 0.2749 -0.0187
-0.7992
2VAL H 11 3.238 1.824 0.826 0.5334 0.1303
0.6497
2VAL CA 12 3.050 1.754 0.901 0.8255 0.0012
0.0684
Then, even if in the .log file (parallel code), the energy
terms are null at every time step to avoid slow
communication between GPU and CPU, the average energies
are null too. (I have test this also using g_energy)
AVERAGES:
Energies (kJ/mol)
Angle Proper Dih. Ryckaert-Bell.
LJ-14 Coulomb-14
0.00000e+00 0.00000e+00 0.00000e+00
0.00000e+00 0.00000e+00
LJ (SR) Coulomb (SR) RF excl.
Potential Kinetic En.
0.00000e+00 0.00000e+00 0.00000e+00
-1.05419e+06 5.24361e+05
Total Energy Temperature Pressure (bar) Cons. rmsd
()
-5.29827e+05 3.39337e+04 0.00000e+00
0.00000e+00
Thanks in advance,
Giulia
On Tue, 13 Apr 2010 19:14:29 +0200
Rossen Apostolov <[email protected]> wrote:
Hi,O
On 04/13/2010 04:23 PM, PACIELLO GIULIA wrote:
Even if I have followed all the instructions reported
for the correct installation of GPU, CUDA, OpenMM and
even if my .mdp file for the simulation is written
following the features supported in the release
Gromacs-OpenMM, the output of the serial code running on
CPU is quite different from that of the parallel code
running on GPU (.gro files).
That is normal. The architectures are different and the
algorithms are different so you can't expect numerically
identical results. The simulated ensembles should be
properly generated though.
Looking at the .log file I have also noticed that the
energy terms 'Angle', 'Proper
Dih.''Ryckaert-Bell.','LJ-14'Coulomb-14', 'LJ (SR)',
'Coulomb (SR), 'RF excl', are null for the simulation
performed with the parallel code.
OpenMM doesn't report individual energy and force terms,
only the totals. This is done for efficiency reasons - it
saves GPU memory and avoids slow CPU<->GPU transfers.
Rossen
--
gmx-users mailing list [email protected]
http://lists.gromacs.org/mailman/listinfo/gmx-users
Please search the archive at
http://www.gromacs.org/search before posting!
Please don't post (un)subscribe requests to the list.
Use the www interface or send it to
[email protected].
Can't post? Read
http://www.gromacs.org/mailing_lists/users.php
--
gmx-users mailing list [email protected]
http://lists.gromacs.org/mailman/listinfo/gmx-users
Please search the archive at http://www.gromacs.org/search before posting!
Please don't post (un)subscribe requests to the list. Use the
www interface or send it to [email protected].
Can't post? Read http://www.gromacs.org/mailing_lists/users.php
