Re: [AMBER] Problem using nmropt=1 with CUDA - AMBER 14

From: David Case <>
Date: Thu, 16 Feb 2017 22:34:57 -0500

On Thu, Feb 16, 2017, Jenner Bonanata wrote:

> Hi, I'm having the following error when running an MD simulation with the
> nmropt=1 option:
> Error: an illegal memory access was encountered launching kernel
> kClearForces cudaFree GpuBuffer::Deallocate failed an illegal memory access
> was encountered

How far into the simulation were you when this error occurred? I'd suggest
(a) making sure that this system runs correctly on CPU's; (b) setting
ntpr to a small value (say 1), to see if you can see problems.

> Here's the restraints file:
> &rst iat=6665,2500, r1=-20.0, r2=1.8, r3=1.8, r4=20.0, rk2=10., rk3=10., /

Note that you are strongly pushing atoms 6665 and 2500 to have a distance
of 1.8 Ang, which is possibly well within the van der Waals repulsive
distance. This will lead to very strong forces on these atoms, which in
turn could cause the integrator to fail. (Things might be OK, if one of these
atoms is a hydrogen.) You could try some experiments to see if one of these
restraints is the culprit; that is, does leaving out just one of the
restraints (rather than turning them all off) give acceptable results?


AMBER mailing list
Received on Thu Feb 16 2017 - 20:00:03 PST
Custom Search