Re: AMBER: Sander problem with different CPU

From: Robert Duke <rduke.email.unc.edu>
Date: Fri, 5 Mar 2004 08:39:37 -0500

Yuann -
You should not be surprised. When sander runs on different numbers of
cpu's, the calculations are done in a different order, because the
calculations are distributed differently. This causes differences in the
rounding errors, and ultimately the results will drift around a bit. The
results you are seeing here are perfectly acceptable, and if you look at the
pairlist totals, the same sorts of pairlists are being generated, the only
difference is they are distributed in the 8 processor case.
I typically see identical output for no more than around 300 steps in MD,
and for some systems, a value or two will vary in far fewer steps in the
last digit, due to the fact that the value as printed out is very near the
"rounding point", and can easily go one way or the other (thus a value
reported as 1.4378 vs 1.4377 may actually be reflecting an internal
difference of 1.437750000000001 vs. 1.437749999999999).
Regards - Bob Duke
----- Original Message -----
From: "yuann" <yuann.bioinfo.ndhu.edu.tw>
To: <amber.scripps.edu>
Sent: Friday, March 05, 2004 7:29 AM
Subject: AMBER: Sander problem with different CPU


> Dear All,
> Recently, I'm supprised that sander gives me different output
> when it applies different number of CPU on IBM P690(AIX51). The
> difference between them initiates at 550steps, and gets larger
> at the NPT equlibrium stage, and also production dynamics.
> System administrator for this machine told me all the AMBER bugfix
> are updated to fix.36.
>
> I'm wondering if this is due to the arrangement of 'SIZE OF NONBOND LIST'
> while running sander with different number of CPU?
>
>
> Here are the output information at the 1st minimization step,
>
>
>
> ---input for sander minimization---
> DNA_drug_complex_WAT 11578atoms 1st_Minimization 1000steps
>
> &cntrl
> imin=1, ntx=1, maxcyc=1000, ncyc=200,
> ntc=1, ntf=1,
> ntb=1, ntp=0,
> ntpr=50,
> cut=9.0,
> ntr=0,
> &end
> &ewald
> &end
>
> --------main difference of output-------
>
> (1cpu):
> | Local SIZE OF NONBOND LIST = 2461494
> | TOTAL SIZE OF NONBOND LIST = 2461494
>
> (8cpu):
> | Local SIZE OF NONBOND LIST = 370628
> | TOTAL SIZE OF NONBOND LIST = 2461494
>
> ----------------------------------------
>
> (1cpu):
> NSTEP ENERGY RMS GMAX NAME NUMBER
> 550 -4.6614E+04 9.7806E-01 2.2339E+01 O5' 543
>
> BOND = 3037.9037 ANGLE = 176.5293 DIHED = 555.3168
> VDWAALS = 7046.6932 EEL = -56278.6230 HBOND = 0.0000
> 1-4 VDW = 341.6471 1-4 EEL = -1493.6799 RESTRAINT = 0.0000
>
> ..........
>
> (8cpu):
> NSTEP ENERGY RMS GMAX NAME NUMBER
> 550 -4.6614E+04 9.7805E-01 2.2339E+01 O5' 543
> BOND = 3037.9031 ANGLE = 176.5292 DIHED = 555.3168
> VDWAALS = 7046.6917 EEL = -56278.6196 HBOND = 0.0000
> 1-4 VDW = 341.6471 1-4 EEL = 1493.6799 RESTRAINT = 0.0000
>
> ..........
> --------------------------------------------------------------------------
-
>
>
> sincerely,
> sychen.
>
> -----------------------------------------------------------------------
> The AMBER Mail Reflector
> To post, send mail to amber.scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
>


-----------------------------------------------------------------------
The AMBER Mail Reflector
To post, send mail to amber.scripps.edu
To unsubscribe, send "unsubscribe amber" to majordomo.scripps.edu
Received on Fri Mar 05 2004 - 13:53:00 PST
Custom Search