Thank you Charles for your answer!
Unfortunately, as I stated in my previous message, running in the CPU for 7
ns the system seems to behave quite fine, there aren't noticeable anomalies
in the energies or physical variables.
I can't understand why only on the GPU is the system explodes.
Regards,
Domenico
On Thu, Apr 28, 2016 at 6:27 PM, Charles Lin <clin92.ucsd.edu> wrote:
> If the CPU simulations are also getting more unfavorable energies with
> each step there may still be something undesirable in your simulation (box
> size, overlap in LJ radius, etc) causing it to explode. My guess (which
> could also be wrong) would be to weaken the restraints you're using. It
> may be that the your molecule strongly prefers being in a distances <r1 or
> >r4 which that portion of the curve follows a linear path so the more it
> deviates out the higher the force and energy contributions will be. A much
> higher force constant can likely cascade into a quick acceleration of your
> system. Preferably for restraints you want the distance to stay within r1
> and r4.
>
> The error itself is likely saying that something in your simulation is
> exploding and its having issues downloading the data off GPU memory to CPU
> memory.
>
> Charlie
>
> ________________________________________
> From: Domenico Marson [domenico87.gmail.com]
> Sent: Thursday, April 28, 2016 3:31 AM
> To: amber.ambermd.org
> Subject: [AMBER] CUDA and restraint
>
> Hello everyone, I'm copying a message I sent one week ago without
> receiving any answer, I know it's a busy time of the year, with the
> upcoming release, probably it went unnoticed!
>
> In the meantime I've been running 7 ns of simulation on CPU and the
> trajectory seems fine both in the data and visualizing it!
>
> Copied text:
>
> Hello everyone,
>
> I'm sorry to bother you just before the release of the "nextgen"
> Amber, but I have some trouble with pmemd.cuda.
>
> I'm trying to run a system with cartesian restraint applied to all the
> atoms of a nanoparticle in explicit TIP3P water,
> while with nmropt I'm restraining the distance of each of 56 different
> atoms on the surface of this nanoparticle.
> To achieve this I restrain the distance between the COM of the
> nanoparticle and each atoms on its surface.
> So, to restrain the distances I have a restraint file with, for N in
> [1, 56] and dist1-4 varying:
> &rst
> iat= -1, N,
> r1=dist1 r2=dist2, r3=dist3, r4=dist4,
> rk2=10.00, rk3=10.00,
> ialtd=0,
>
> igr1=1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16,17,18,19,20,21,22,23,24,25,26,27,28,29,30,31,32,33,34,35,36,37,38,39,40,41,42,43,44,45,46,47,48,49,50,51,52,53,54,55,56,57,58,59,60,61,62,63,64,65,66,67,68,69,70,71,72,73,74,75,76,77,78,79,80,81,82,83,84,85,86,87,88,89,90,91,92,93,94,95,96,97,98,99,100,101,102,103,104,105,106,107,108,109,110,111,112,113,114,115,116,117,118,119,120,121,122,123,124,125,126,127,128,129,130,131,132,133,134,135,136,137,138,139,140,141,142,143,144,145,146,147,
> /
>
> I have amber patched (compiled) with all the patches available as
> today, and my GPU is a Tesla K20c with Driver Version 352.79 and cuda
> 7.5.
>
> I performed minimisation, heating to 300 K (20 ps) and a fist
> equilibration of density (50 ps) on the CPU without any problem.
> Energy are fine and also the trajectory/behaviour seems fine.
> Than I wanted to continue on the GPU but, no matter what combination
> of ntc=2 + ntf=1 or ntc=2 + ntf=2, my simulation blows up just in a
> few steps with error:
> "cudaMemcpy GpuBuffer::Download failed an illegal memory access was
> encountered".
> I tried to output every frames in the mdout and trajectory, but I
> can't see the reason why It's blowing up.
> I see only my nanoparticle "exploding" step by step, and the restraint
> and bond energies increasing to reach "******" in the first 5-10
> steps.
> Moreover, If I continue with the same settings on the CPU no problem
> arises (at least not in a reasonable time, I have only 6 cores
> available).
>
> I know many patches were released for COM restraint on GPU, maybe
> something else is missing? Or I'm just trying too much?
> Thank you all for your help!
>
> Regards,
> Domenico
>
> --
> Domenico Marson, Ph.D.
> Department of Engineering and Architecture (DEA) Postdoctoral Fellow
> Molecular Simulation Engineering (MOSE) Laboratory
>
> University of Trieste
>
> Skype: domenicomars
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
> _______________________________________________
> AMBER mailing list
> AMBER.ambermd.org
> http://lists.ambermd.org/mailman/listinfo/amber
>
--
*Domenico Marson, Ph.D.*
Department of Engineering and Architecture (DEA) Postdoctoral Fellow
Molecular Simulation Engineering (MOSE) Laboratory
University of Trieste
Skype: domenicomars
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Apr 29 2016 - 01:30:10 PDT
