AMBER Archive (2004)

Subject: Re: AMBER: Computing psec/day using Amber 8

From: David A. Case (case_at_scripps.edu)
Date: Thu May 06 2004 - 14:27:10 CDT


On Thu, May 06, 2004, Craig TOEPFER wrote:
>
> First, what time do I use to compute seconds per step. Do I use the
> total runtime(i.e., time pmemd....) or some subset of the total time?
> At the end of the output file, I see a breakdown of time spent per
> component. Do I use one of these?

We generally use the "non-setup wallclock time".

>
> Finally, does the value of 86.4 only apply to the "jac" benchmark, or is
> this a generic value? I saw on the Amber 8 benchmarks page that the
> value for the "factor_ix" benchmark is 129.6, not 86.4.

This is not a "magic" conversion. There are 24 x 60 x 60 = 86,400 seconds
in a day. If the code takes n second per fs, you would get 86.4/n psec
done per day.

As the web page states, "[fix] uses dt=0.0015, so the conversion is
ps-per-day = 129.6/(time-per-step)": the conversion factor is dependent on
both the time-per-step and on the step size.

...dac

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