[AMBER] CUDA and restraint

From: Domenico Marson <domenico87.gmail.com>
Date: Thu, 28 Apr 2016 12:31:34 +0200

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
Received on Thu Apr 28 2016 - 04:00:04 PDT
Custom Search