Re: [AMBER] ff19SB with end-caps pmemd.cuda problem

From: Carlos Simmerling <>
Date: Wed, 11 Dec 2019 06:53:44 -0500

We haven't seen problems with standard caps - they actually use ff14SB so
should not be a problem. There are times when a poor quality structure may
just be ok with Cuda for ff19SB but not ff14SB (or vice versa) - is this
for a well equilibrated system?

On Wed, Dec 11, 2019, 5:38 AM Pavlína Pokorná <>

> Dear all,
> when attempting simulations with ff19SB with the updated Amber18 version
> from Carlos Simmerling's GitHub, I failed to run pmemd.cuda simulations
> with ACE and NHE caps on N- and C- termini of the protein. Simulations
> crashed with the following error in ~50 nanoseconds: "Error: an illegal
> memory access was encountered launching kernel kNLSkinTest". No further
> error message is printed in the simulation *.out file and the last-saved
> structure showed no abnormalities. Simulations with the same system without
> the end-caps run without any problems. Runs with the end-caps with sander
> on CPUs do not result in this error. Test ff14SB cuda simulations with the
> end-caps with the updated Amber18 run also without problems. I have ~1100
> atoms in my system.
> It seems that cuda + ff19SB + caps combination produces some large numbers
> somewhere. Is there something I am missing or could you advise any possible
> solution to my problem?
> The ff19SB improved some local backbone conformations (and associated
> protein-protein interactions) wrt ff14SB in my protein-protein complex, so
> I'd like to use this force field. Still, I need to use the end-caps due to
> the excessive and undesired end-end interactions in the structure.
> I'd be grateful for any comment.
> Best wishes,
> Pavlina
> --
> Pavlina Pokorna
> *PhD student*
> Institute of Biophysics of the CAS, v. v. i.
> Kralovopolska 135, 612 65 Brno
> _______________________________________________
> AMBER mailing list
AMBER mailing list
Received on Wed Dec 11 2019 - 04:00:02 PST
Custom Search