Project

General

Profile

Bug #1134

gromacs 4.6 GB/SA problem and poor performance

Added by Changwon Yang over 6 years ago. Updated over 5 years ago.

Status:
Closed
Priority:
Normal
Assignee:
Category:
-
Target version:
Affected version - extra info:
Affected version:
Difficulty:
uncategorized
Close

Description

Im trying to run an md or em using an implicit solvation method using
gromacs 4.6 but I always get the incorrect result.
ICC version : icc 11.0
fftw version : 3.2.2

benchmark system is gromacs-gpubench
gromacs-gpubench-dhfr.tar/CPU/dhfr-impl-inf.bench

Angle,Proper Dih,Imp Dih,Nonpolar sol,LJ-14,Coulomb-14 energy are correct.
but GB polarization energy is too low, LJ,Coulomb(SR) energy are always
zero.

It seems that there is a bug in the program.

Using gromacs 4.5. It works fine.

gromacs 4.5.3 : 9.4ns/day
gromacs 4.6 :2.4ns/day

Input files: conf.gro and mdp files

http://www.gromacs.org/Documentation/Installation_Instructions_4.5/GROMACS-OpenMM

cpu-imp-RF-inf.mdp

constraints = all-bonds
integrator = md
dt = 0.002 ; ps !
nsteps = 0
nstlist = 0
ns_type = grid
rlist = 0
coulombtype = cut-off
vdwtype = cut-off
rcoulomb = 0
rvdw = 0
pbc = no
epsilon_rf = 0
rgbradii = 0
comm_mode = angular

implicit_solvent = GBSA
gb_algorithm = OBC
gb_epsilon_solvent = 78.3
sa_surface_tension = 2.25936

nstxout = 0
nstfout = 0
nstvout = 0
nstxtcout = 50
nstlog = 0
nstcalcenergy = -1
nstenergy = 50

tcoupl = berendsen
tc-grps = system
tau-t = 0.1
ref-t = 300

gromacs 4.6 (icc 11.0, centos 5.8, fftw 3.2.2)

Statistics over 101 steps [ 0.0000 through 0.2000 ps ], 10 data sets
All statistics are over 11 points

Energy Average Err.Est. RMSD Tot-Drift
-------------------------------------------------------------------------------
Angle 4189.86 -- 1094.25 2527.08 (kJ/mol)
Proper Dih. 6600.66 -- 234.905 471.156 (kJ/mol)
Improper Dih. 121.598 -- 32.7574 64.572 (kJ/mol)
GB Polarization 32708.8 - 1979.66 6416.79 (kJ/mol)
Nonpolar Sol. 158.323 -
13.0032 42.232 (kJ/mol)
LJ-14 2346.66 -- 180.767 289.567 (kJ/mol)
Coulomb-14 26555.4 -- 1091.98 3000.88 (kJ/mol)
LJ (SR) 0 -
0 0 (kJ/mol)
Coulomb (SR) 0 -- 0 0 (kJ/mol)
Potential 7263.71 -- 1835.42 -6023.06 (kJ/mol)

gromacs 4.5.3

Statistics over 101 steps [ 0.0000 through 0.2000 ps ], 10 data sets
All statistics are over 11 points

Energy Average Err.Est. RMSD Tot-Drift
-------------------------------------------------------------------------------
Angle 1840.12 -- 178.343 561.256 (kJ/mol)
Proper Dih. 6207 -- 101.135 210.593 (kJ/mol)
Improper Dih. 68.4513 -- 11.6214 29.5806 (kJ/mol)
GB Polarization 13534.1 - 34.3666 95.8711 (kJ/mol)
Nonpolar Sol. 143.048 -- 1.55868 5.13589 (kJ/mol)
LJ-14 2380.44 -- 26.9563 24.6177 (kJ/mol)
Coulomb-14 29155.5 -- 64.9054 176.267 (kJ/mol)
LJ (SR) -5274.28 -
214.048 650.865 (kJ/mol)
Coulomb (SR) 38297.8 - 143.656 398.004 (kJ/mol)
Potential -17311.7 -
451.581 1003.65 (kJ/mol)

topol.top (708 KB) topol.top Changwon Yang, 01/27/2013 12:26 PM
cpu-imp-RF-inf.mdp (819 Bytes) cpu-imp-RF-inf.mdp Changwon Yang, 01/27/2013 12:26 PM
conf.gro (109 KB) conf.gro Changwon Yang, 01/27/2013 12:26 PM
out4.5.3 (1.02 KB) out4.5.3 Changwon Yang, 01/27/2013 12:26 PM
out4.6 (1.02 KB) out4.6 Changwon Yang, 01/27/2013 12:26 PM

Related issues

Related to GROMACS - Feature #1095: Fix all-vs-all kernelsClosed
Related to GROMACS - Bug #1249: no-PBC no-cutoff is broken with SIMD group kernels in 4.6.1Closed05/14/2013

History

#1 Updated by Changwon Yang over 6 years ago

Using inf.cut-off, I always get the incorrect result. using finite cut-off, It works fine.(ICC version 13.0 or 11.0).

#2 Updated by Mark Abraham about 6 years ago

  • Target version set to 4.6.x
  • Affected version set to 4.6

Thanks. This is a known bug. See http://redmine.gromacs.org/issues/1249. Setting GMX_NOOPTIMIZEDKERNELS gives correct results. The Verlet scheme also gives correct results.

#3 Updated by Mark Abraham about 6 years ago

  • Status changed from New to In Progress
  • Assignee set to Erik Lindahl

#4 Updated by Rossen Apostolov over 5 years ago

  • Status changed from In Progress to Resolved

Closing that too

#5 Updated by Rossen Apostolov over 5 years ago

  • Status changed from Resolved to Closed

Also available in: Atom PDF