Re: [AMBER] problem of GPU with ibelly

From: Hadi Rahmaninejad <>
Date: Wed, 7 Nov 2018 18:59:19 -0500

you are right, but I am not putting ibelly on all of my atoms. just need to fixed the atom in a lattice. I tried ntr=1 but since I did not defined bonded forces beetween atoms in structure, the lattice expanded. My problem now is that if I run with cpu, the simulation is unreasonably slow; for about 50000 atom, it is just 0.37 ns/day. I appreciate if you give me any idea for that also,

Sent from my iPad

> On Nov 7, 2018, at 2:50 PM, David A Case <> wrote:
>> On Wed, Nov 07, 2018, Hadi Rahmaninejad wrote:
>> Du to my long simulation, I need to run my job using GPU instead of CPU.
>> However, since I am "using ibelly, I get this error when I run with gpu:
>> Implementation does not support Belly options.
>> Require ibelly == 0.
>> Input errors occurred. Terminating execution."
> Could you use restraints (ntr=1) rather than a belly (ibelly=1)?
> A "belly" calculation interupts Newton's equations of motion by
> artifically setting forces on certain atoms to zero (at one particular
> point in the integration scheme.) This can do bad things to thermodynamic
> quantities like temperature, pressure, and to other parts of the
> integrator (like SHAKE). Having a moving piece of the system in contact
> with the fixed piece can lead to imbalances: the behavior (say) of water
> around a "rigid" but still fluctuating entity might be different than its
> behavior around a completely fixed entity. Further, the implementation in
> pmemd doesn't even save you any time.
> Basically, there has been little impetus for developers to add this "feature"
> to pmemd.cuda. So it is unlikely that this will get added any time soon (but
> one never knows, since someone else might want this option as well, and know
> how to implement it.)
> ....dac
> _______________________________________________
> AMBER mailing list

AMBER mailing list
Received on Wed Nov 07 2018 - 16:00:02 PST
Custom Search