Hi Hoshin,
You are probably hitting some assumption we made in the GPU code.
Certainly I've never tried doing simulations of Gold surfaces with it and
that is way outside the scope of what most people would do.
Firstly, please confirm that you are using AMBER 14 with all the latest
patches. Look in your mdout file for the section that begins:
--------------------- INFORMATION ----------------------
| GPU (CUDA) Version of PMEMD in use: NVIDIA GPU IN USE.
And paste in the reported version and date here.
Next try running your heating and initial equilibration on the CPU and
then switch to the GPU and see if that helps.
Finally confirm that if you use the exact same input with the exact same
random seed (set ig explicitly) that the situation yields NANs at exactly
the same point. This is critical and will establish that it is a bug in
the code and not a misbehaving GPU.
Once you have done this and have a reproducible test case that shows this
behavior please post it here and we can try to figure out what the problem
is.
All the best
Ross
On 8/4/14, 12:04 PM, "Hoshin Kim" <85hskim.gmail.com> wrote:
>Dear all,
>
>I am doing MD simulations of DNA grafted on Au surface. For simulations,
>Amber GPU computing system are being used (Exxact, GTX 780)
>
>Now I am having a hard time performing MD simulations because of following
>error:
>
>When I do MD simulation (I've tried both NVT, and NPT conditions), all
>information in md.restrt and some terms in mdinfo (md.out) suddenly turned
>into NaN, and NaN with asterisks, respectively, at random time step.
>To figure this problem, I took a restrt file right before error occurred,
>and reran MD. It worked fine first time, but identical error occurred
>again
>at random time step.
>
>Here is an example of mdinfo file
> NSTEP = 49999500 TIME(PS) = 100039.000 TEMP(K) = NaN PRESS =
>0.0
> Etot = NaN EKtot = NaN EPtot =
>**************
> BOND = 0.0000 ANGLE = 70230.8688 DIHED =
>0.0000
> 1-4 NB = 0.0000 1-4 EEL = 0.0000 VDWAALS =
>**************
> EELEC = ************** EHBOND = 0.0000 RESTRAINT =
>0.0000
>
>--------------------------------------------------------------------------
>----
>
>Interestingly, no errors were observed when I did same MD simulations
>using
>CPU, instead of GPU. Plus, For more simple conditions (e.g. just DNA in
>water) using same input parameters for minimization, heating, and
>production run, I had no problems using GPU.
>
>Regards,
>
>Hoshin
>_______________________________________________
>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 04 2014 - 12:30:03 PDT
