I would not necessarily trust the installation.
Run the minimization is serial on PMEMD. Then switch to running MD with the
GPU code in serial and, assuming the system you are simulating is good, you
should be ok to run. If it fails with pmemd in serial then there is
something very wrong with your underlying prmtop and inpcrd files.
BTW, for good measure you should probably build your own version of AMBER on
the machine to make sure it has the very latest bug fixes. This is
especially true if you are using the GPU code to simulate lipids - the
minimum patch level for AMBER 12 should be bugfix.18 you can check this
with:
cd $AMBERHOME
./updated_amber --show-applied-patches
If this doesn't show the code patched all the way to bug fix.18 (which
should report the GPU code as being v12.3 in mdout when you run it) and you
still want to use the provided AMBER 12 module then contact the admins at
TACC and have them update everything before proceeding further.
All the best
Ross
From: Rejwan <rejwan_88.yahoo.com>
Reply-To: Rejwan <rejwan_88.yahoo.com>
Date: Monday, August 19, 2013 1:58 PM
To: Ross Walker <ross.rosswalker.co.uk>, AMBER Mailing List
<amber.ambermd.org>
Cc: Dave Case <case.biomaps.rutgers.edu>
Subject: Re: [AMBER] AMBER minimization job
Hi
Thank you. I'm bit puzzled as same minimization run (same topology and
inpcrd) are reporting different number of non-bonded pair and non-bonded
list in the GPU queue. As before minimization could not complete but then
why different NB pair and list ? Thank you for your input. Regards.
-Rejwan
_______________________
TACC: Done.
TACC: Starting up job 1365319
TACC: Setting up parallel environment for MVAPICH ssh-based mpirun.
TACC: Setup complete. Running job script.
TACC: starting parallel tasks...
* NB pairs 358 6860550 exceeds capacity ( 6860880) 1
SIZE OF NONBOND LIST = 6860880
Exit code -5 signaled from c300-204
* NB pairs 271 6860766 exceeds capacity ( 6860880) 3
SIZE OF NONBOND LIST = 6860880
SANDER BOMB in subroutine nonbond_list
Non bond list overflow!
From: Ross Walker <ross.rosswalker.co.uk>
To: Rejwan <rejwan_88.yahoo.com>; AMBER Mailing List <amber.ambermd.org>
Sent: Monday, August 19, 2013 4:18 PM
Subject: Re: [AMBER] AMBER minimization job
Hi Rejwan,
I suspect that something is wrong with your initial structure then -
perhaps the density is way too high, or the box size is completely wrong.
Something is leading to you having a VERY large pair list - I would
suspect the density is crazy high. How did you build your structure? -
Have you visualized it and checked it looks ok? Looked at it in VMD? Check
the bond lengths etc?
Try running it in serial first using just pmemd. It is possible the
installation of AMBER 12 on Stampede was not done right and or tested.
Also note you are asking for a VERY large number of cores:
#$ -pe 2way 180
There is little to no chance this will run efficiently with sander,
probably not with pmemd either - I'm not surprised sander just barfs, it's
unlikely to have been tested with more than 64 cores at most. PMEMD should
be fine at such a core count although I doubt you'll get very good
performance - especially for minimization which is far from optimized in
parallel.
Where did you come up with this core count from?
All the best
Ross
On 8/19/13 1:00 PM, "Rejwan" <rejwan_88.yahoo.com> wrote:
>Hi,
>
>I am trying to simulate a GPCR in POPC bi-layer. The minimization
>protocol is mainly based on following lipid tutorial link.
>
>http://ambermd.org/tutorials/advanced/tutorial16/An_Amber_Lipid_Force_Fiel
>d_Tutorial.html
>
>
>Minimize the lipid structure
> &cntrl
> imin=1,
> maxcyc=10000,
> ncyc=5000,
> ntb=1,
> ntp=0,
> ntf=1,
> ntc=1,
> cut=10.0,
> ntpr=50,
> ntwr=2000,
> &end
>/
>
>I'll update you on PMEMD run. However is there way to reduce/ignore
>non-bonded terms ? Thanks.
>
>Rejwan
>
>
>________________________________
> From: David A Case <case.biomaps.rutgers.edu>
>To: Rejwan <rejwan_88.yahoo.com>; AMBER Mailing List <amber.ambermd.org>
>Sent: Monday, August 19, 2013 3:31 PM
>Subject: Re: [AMBER] AMBER minimization job
>
>
>On Mon, Aug 19, 2013, Rejwan wrote:
>>
>> An AMBER minimization job was aborted due to overflow of non-bonded
>> list. My system is of 52 K atoms size. How I can perform minimization
>> in such situation ? Thank you.
>>
>>
>> Exit code -5 signaled from c321-214
>> * NB pairs 522 1219601 exceeds capacity ( 1219712) 23
>> SIZE OF NONBOND LIST = 1219712
>
>Can you post your min_lipid.in file? Also, try a short serial
>(non-parallel)
>run to see if the problem is specific to the parallel code. Trying pmemd
>instead of sander is also worth trying, since the nonbond lists are
>handled
>in a different way there.
>
>....dac
>
>
>_______________________________________________
>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 Mon Aug 19 2013 - 14:30:15 PDT
