First, your init.pdb is really wacky. You've got a bleb in the membrane. That appears to heal rapidly during simulation, but the Z-axis fluctuations look pretty large to me. I suspect you're running into a known problem with P-R pressure coupling, which is that oscillations can get out of hand if your run start far from equilibrium (which yours clearly does). The solution is to run for a while first with Berendsen pressure coupling to get the system close to equilibrium and then switch over to P-R for production.
I suspect that if you do Berendsen for 100 ns then your P-R simulation will be stable on the GPUs as well. I can not answer any question as to why you seem to be having trouble on GPUs and not CPUs, and perhaps your simulation has uncovered a bug somewhere, but for now I'd suggest that bad inputs can always yield bad outputs with some probability. If you're still searching for the difference, then I'd look at mdrun -notunepme, since that would be one actual difference between CPU and GPU runs. Also, you're not using the charmm force field properly with your .mdp file. That could be beside the point, but nstlist = 20 with rlist = 1.2 and your other setting is not normal, right? Try nstlist = 10. I realize that gromacs is going to mess with this during optimization, but I presume that it starts with what you give it. Thank you, Chris. -- Gromacs Users mailing list * Please search the archive at http://www.gromacs.org/Support/Mailing_Lists/GMX-Users_List before posting! * Can't post? Read http://www.gromacs.org/Support/Mailing_Lists * For (un)subscribe requests visit https://maillist.sys.kth.se/mailman/listinfo/gromacs.org_gmx-users or send a mail to gmx-users-requ...@gromacs.org.