Project

General

Profile

Bug #1299

Updated by Mark Abraham over 7 years ago

From http://gromacs.5086.x6.nabble.com/Inconsistent-results-between-3-3-3-and-4-6-with-various-set-up-options-tp5009584p5009585.html

Mark said I should open an issue to see if the bad performance, atypical RSMD, and zero potentials (at step 0) I get when running 4.6 on GPU are reproducible.

Temperature Potential E. Pressure
System name Details Average RMSD Average RMSD Average RMSD ns/day
3.3.3 c_md RF nst5 group 306.0 1.4 -439029 466 0.998 125 9.7
4.6 c_md RF nst5 group 303.9 1.4 -440455 461 0.0570 126 36.9
4.6 c_vv RF nst5 verlet 303.0 1.2 -438718 478 1.96 134 16.5
4.6 g_md RF nst20 verlet 303.0 1.4 -439359 3193 566 1139 19.8
4.6 g_vv RF nst20 verlet 303.0 1.2 -438635 3048 34.3 405 18.6
4.6 c_pme md nst5 group 303.0 1.4 -436138 461 0.135 125 25.5
4.6 g_pme md nst40 verlet 303.0 1.4 -431621 463 416 1016 33.6

Where c_md indicates CPU only and md integrator, g_vv indicates GPU and md-vv integrator, etc. Verlet & group refer to cut-off scheme and nst# refers to nstlist frequency which was automatically changed by gromacs. I found very similar results (and run times) for the GPU runs when -nb was set to gpu or gpu_cpu.

<pre>Step


Step
0:
System LJ (SR) Coulomb (SR) Potential Kinetic En. Total Energy Temperature Pressure (bar)
3.3.3 c_md 1.80072E+04 -4.30514E+05 -4.38922E+05 6.14932E+04 -3.77429E+05 3.06083E+02 1.53992E+02
4.6 c_md 1.80072E+04 -4.30515E+05 -4.38922E+05 6.20484E+04 -3.76874E+05 3.08847E+02 1.56245E+02
4.6 c_vv 1.15784E+04 -4.83639E+05 -4.37388E+05 6.14748E+04 -3.75913E+05 3.05992E+02 -1.40193E+03
4.6 g_md 0.00000E+00 0.00000E+00 3.46728E+04 6.14991E+04 9.61719E+04 3.06113E+02 -1.70102E+04
4.6 g_vv 0.00000E+00 0.00000E+00 3.46728E+04 6.14748E+04 9.61476E+04 3.05992E+02 -1.85758E+04
4.6 c_pme 1.30512E+04 -3.37973E+05 -4.35821E+05 6.14989E+04 -3.74322E+05 3.06112E+02 4.50028E+02
4.6 g_pme 1.76523E+04 -4.89006E+05 -4.31207E+05 6.14990E+04 -3.69708E+05 3.06112E+02 4.37951E+02</pre> 4.37951E+02

Back