Thanks. Tried that as @MagnusL suggested above already which resulted in the same error reported 5 minutes ago.
My apologies, the previous test with changed barostat (c-rescale), tau_t (1) and tau_p (5) ended in the same error 719, not as I first reported (700, forgot to rerun grompp). I have revised the above message.
I tried extending both NVT and NPT from 100 ps to 10 ns. Both completed without issues. I next tried 100-ns MD (original md.mdp parameters), still got cuda error 719. I then reran an extended NPT to 100 ns (with your suggested modifications, i.e. c-rescale, tau_t=1,1; tau_p=5). The same cuda error occurred after 20 ns. So, it happens during equilibration and under positional restraints.
Thank you @hess@MagnusL for ideas. I think I’ll stop here for the time being. Issues remain unresolved.
This happened to me a couple of times when I was working on an HPC node, and several other MD simulation programs (amber (pmemd.cuda) and desmond) were running on the same node. Can this cause the problem?