Re: AMBER: A few questions in the protein unfolding simulations using GB model

From: Jiayun Pang <JXP251.bham.ac.uk>
Date: Thu, 29 Jan 2004 10:47:52 -0000

Hi, thanks to all the answers.

Carlos, you mentioned it's not a good idea to use cutoff in GB. I read the
GB tutorial in the amber website. Ideally, one should use no cut-off , but
as I found in my case, it is too computationally expensive. In the previous
discussion in the amber reflector, it is said larger values, such as 15 to
24 A could be considered in GB. I assume, as it is said in the tutorial ,
it is the trade between the speed and accuracy of the simulations. Do you
have any suggestions?

I am using Amber7 with the mpi run before. As you suggested, I have a
shorter run with single processor. The dimer run is 4.2 times longer the
monomer run. It seems the time allocation among GB parameters are different.

The following is the detailed analysis at the end of the output file for the
Dimer run.
----------------------------------------------------------------------------

----
   5.  TIMINGS
----------------------------------------------------------------------------
----
|                Calc gb radii            683.05 (19.52% of Gen B)
|                Calc gb diag            2211.56 (63.20% of Gen B)
|                Calc gb off-diag         465.40 (13.30% of Gen B)
|                Surface area energy      139.50 ( 3.99% of Gen B)
|             Gen Born time           3499.51 (100.0% of Nonbo)
|          Nonbond force           3499.51 (99.40% of Force)
|          Bond energy                0.39 ( 0.01% of Force)
|          Angle energy               4.57 ( 0.13% of Force)
|          Dihedral energy           16.30 ( 0.46% of Force)
|       Force time              3520.77 (99.86% of Runmd)
|       Shake time                 2.74 ( 0.08% of Runmd)
|       Verlet update time         2.17 ( 0.06% of Runmd)
|    Runmd Time              3525.68 (100.0% of Total)
| Total time              3387.34 (100.0% of ALL  )
| Highest rstack allocated:      58894
| Highest istack allocated:      21416
|     Setup wallclock           1 seconds
|     Nonsetup wallclock     3387 seconds
The Monomer output is,
----------------------------------------------------------------------------
----
   5.  TIMINGS
----------------------------------------------------------------------------
----
|                Calc gb radii            274.64 (31.91% of Gen B)
|                Calc gb diag             347.72 (40.40% of Gen B)
|                Calc gb off-diag         171.48 (19.92% of Gen B)
|                Surface area energy       66.88 ( 7.77% of Gen B)
|             Gen Born time            860.72 (100.0% of Nonbo)
|          Nonbond force            860.72 (98.87% of Force)
|          Bond energy                0.14 ( 0.02% of Force)
|          Angle energy               1.98 ( 0.23% of Force)
|          Dihedral energy            7.68 ( 0.88% of Force)
|       Force time               870.52 (99.83% of Runmd)
|       Shake time                 0.71 ( 0.08% of Runmd)
|       Verlet update time         0.78 ( 0.09% of Runmd)
|    Runmd Time               872.01 (100.0% of Total)
| Total time               805.65 (100.0% of ALL  )
| Highest rstack allocated:      29447
| Highest istack allocated:      10708
|     Setup wallclock           0 seconds
|     Nonsetup wallclock      805 seconds
----- Original Message ----- 
From: "Carlos Simmerling" <carlos.csb.sunysb.edu>
To: <amber.scripps.edu>
Sent: Wednesday, January 28, 2004 6:26 PM
Subject: Re: AMBER: A few questions in the protein unfolding simulations
using GB model
> I would suggest taking a closer look at the detailed timings provided at
the
> end of the sander output, and running single cpu so as not to confuse
> the issue even more with different scaling efficiencies for the systems.
>
> GB certainly doesn't scale linearly, but you're also using a cutoff
(perhaps
> not a good idea) so it's not be N^2 as someone suggested.
> also GB doesn't use a pairlist, so it isn't pairlist generation as Bill
> suggested.
>
> run shorter benchmarks to test this, you should get reliable timings
> with a 15 minute or so run, no need to run for days.
>
> which amber version is this?
>
> carlos
>
>
> -----------------------------------------------------------------------
> 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 Jan 30 2004 - 18:56:03 PST