AMBER Archive (2005)

Subject: Re: AMBER: leap single precision

From: David A. Case (case_at_scripps.edu)
Date: Sun Apr 10 2005 - 00:27:52 CDT


On Fri, Apr 08, 2005, Guanglei Cui wrote:

> My impression is that the maximum number of particles that a double
> precision leap can handle is around 2 to 3 million (8Gbyte memory).

Note that the file formats used downstream (for example, by sander) have a
limit of one million atoms. So there is a lot that would have to be done
to handle systems of the size you are thinking of.

Of the test cases we currently use, the largest has about 140,000 atoms.
Mike Crowley is preparing a test case of about 700,000 atoms, so we will try
to make sure that Amber 9 works (efficiently) at that size of system.

But, going beyond that is currently a (non-trivial) research problem.

A note about the problem of reading frcmod files with a single-precision LEaP:
the dFortranReadDouble() routine is what reads input from these sorts of files.
>From the problem description, I would bet money that you need to create a
dFortranReadSingle() routine if you want the results to be placed in a
single-precision float rather than a double. This would involve changing:

    sscanf( sDouble, "%lf", &dVal );

with

    sscanf( sDouble, "%f", &fVal );

where "fVal" is a float and dVal is a double.

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