AMBER Archive (2006)

Subject: Re: AMBER: Different results between serial and parallel runs in AMBER 8

From: Ilyas Yildirim (yildirim_at_pas.rochester.edu)
Date: Tue Mar 07 2006 - 06:51:14 CST


Dear DAC,

I followed your advise, and did 3 MD runs; 1 serial and 2 parallel runs
with different CPUs (one with 2 CPUs and second with 4 CPUs). I used a
.rst file to start with (which has velocity info. in it), and here is the
md.in file I used in all cases:

------------------------------- md.in -------------------------------
    10 fs simulation
     &cntrl
      imin=0,
      ntx=5,irest=1,
      ntpr=1,ntwr=1,ntwx=1,
      ntc=2,ntf=2,ntb=1,cut=8,
      igb=0,
      ntr=0,
      nstlim=10,dt=0.001,nscm=5000,nrespa=1,
      ntt=3,gamma_ln=1,tempi=300,temp0=300,ig=233,
      ntp=0,taup=2.0,pres0=1
     /
----------------------------------------------------------------------

Indeed, all of the runs gave exactly the same values: Everything
(including the TEMP(K) values) is same. I will change ntt to 1 as
suggested by Carlos to see if ntt=3 case is creating this. But it
is mind bugging to see different results when inpcrd does not have any
velocity info in it. Thanks.

Best,

On Mon, 6 Mar 2006, David A. Case wrote:

> On Mon, Mar 06, 2006, Ilyas Yildirim wrote:
> >
> > I have a question on parallel and serial MD runs in AMBER 8.
>
> It would be better to compare restart runs (irest=1) rather than runs
> where random numbers are used to get initial temperature. However,
> even the tempi stuff *should* be giving the same results for different numbers
> of processors. There may be a funny interaction between ntt=3 and tempi:
> it looks like tempi "uses up" some random numbers on the master node,
> but these are not equivalently cycled through on the slave nodes. So, then
> the random numbers for Langevin will not be the same on the master and slave
> nodes....setting irest=1 will avoid this "problem".
>
> Let me know what happens; this buglet appears to still be there in amber9.
>
> ...dac
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>
>

-- 
  Ilyas Yildirim
  ---------------------------------------------------------------
  - Department of Chemisty       -				-
  - University of Rochester      -				-
  - Hutchison Hall, # B10        -				-
  - Rochester, NY 14627-0216     - Ph.:(585) 275 67 66 (Office)	-
  - http://www.pas.rochester.edu/~yildirim/			-
  ---------------------------------------------------------------

----------------------------------------------------------------------- The AMBER Mail Reflector To post, send mail to amber_at_scripps.edu To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu