On Sun, Mar 07, 2021, Ming Tang wrote:
>
>Thanks for the precious guidance. This what I want and the script was
>perfect. But I after 4000 steps the production run got this error:
>cudaMemcpy GpuBuffer::Download failed an illegal memory access was
>encountered. And there is not restrain in the output as below:
>
>wrapping first mol.: 30.51583 43.15591 74.74821
>wrapping first mol.: 30.51583 43.15591 74.74821
>
> NSTEP = 4000 TIME(PS) = 608.000 TEMP(K) = 308.07 PRESS = -237.6
> Etot = -153810.1794 EKtot = 36112.7344 EPtot = -189922.9138
> BOND = 578.5793 ANGLE = 1445.0337 DIHED = 2202.2790
> 1-4 NB = 687.3814 1-4 EEL = 7970.8282 VDWAALS = 24116.2997
> EELEC = -226923.3149 EHBOND = 0.0000 RESTRAINT = 0.0000
> EKCMT = 17106.3674 VIRIAL = 20136.3763 VOLUME = 590645.1967
> Density = 0.9997
> ------------------------------------------------------------------------------
>
> NMR restraints: Bond = 0.000 Angle = 0.000 Torsion = 0.000
A zero restraint energy probably just means that all the restraints are
satisfied, that is, that the average distance being restrained is greater
than 1.8 Å.
>
>&rst iat=-1,-1,0
> r1=1.300000, r2=1.800000, r3=10.000000, r4=10.500000, rk2=20.000000, rk3=0.000000,
> IGR1(1)=2794,IGR1(2)=2795,IGR1(3)=2796,...
> IGR2(1)=2829,IGR2(2)=2830,IGR2(3)=2831,...
> nstep1=0, nstep2=0,
> &end
Since you set rk3=0, this is just a lower bound. That is fine if it what
you want.
> nstlim=2000000000,dt=0.002,
Debugging steps:
1. Since the restraints aren't doing anything anyway, run a short
simulation (say 10000 steps, not 2 billion!) and see if you still get the
crash.
2. Set iwrap=0 (you should be doing this anyway). It's certainly possible
that the restraints get confused when wrapping is going on.
3. If the problem seems to be with the restraints, try a test run with many
fewer elements in the IGR1, IGR2 arrays. It's not clear how big those
arrays are, but there may be some upper bound.
4. Try setting ntb=1, ntp=0 to see if there is something odd about pressure.
> ntpr=1000, ntwx=1000,ntwr=1000,iwrap=1,
Note that if this job actually ran, you would probably exhaust all disk
space: you are asking to dump a trajectory every 1000 steps, so you would
get 2000000000/1000 = 2 million frames in your trajectory file(!)
...good luck...dac
_______________________________________________
AMBER mailing list
AMBER.ambermd.org
http://lists.ambermd.org/mailman/listinfo/amber
Received on Sun Mar 07 2021 - 05:30:03 PST
