Does it blow up at the same place each time on the GPU? E.g. tail the
.out files.
Bill
On 4/29/16 1:19 AM, Domenico Marson wrote:
> 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
>>
>
>
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Fri Apr 29 2016 - 01:30:11 PDT
