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
Received on Thu Apr 28 2016 - 09:30:03 PDT
