AMBER Archive (2003)

Subject: Re: Simple math of memory calculation for nmode

From: David A. Case (case_at_scripps.edu)
Date: Thu May 01 2003 - 19:23:06 CDT


On Thu, May 01, 2003, Masaki Tomimoto wrote:
>
> I have a question about a simle mathematics to calculate how much memory I
> need. I found the following comments from Dr. Case in Amber Mailing
> Archive.
>
>
> My system contains about 4000 atoms. According to the above math, I should
> need (4000x3)x(4000x3)/2 = 144,000,000 words, or 1.15 GB. However, when I
> execute nmode, it says that it require almost double, 208,497,233. Even
> Hessian is a symmetric matrix which requires only triangle part (that
> denotes "/2" is the above math), nmode requires memory space which hold
> entire hessian matrix?
>

The current version of nmode actually requires about 3*(3n)**2/2 words of
memory for the matrices. This would be 216 million words, or 1.7 Gbytes for a
4000 atom system; some smaller amounts of memory are needed for other tasks.
You can follow the allocation by looking at the "ntrun.eq.1" sections of
alloc.f.

This is a tradeoff of memory usage for faster computational time. I would be
possible to use less than half this amount of memory, and that may be an
important consideration for clusters that have limited amounts of physical
memory. I can work off-line with you to put together such a code if that
seems necessary.

..dave case

-- 

================================================================== David A. Case | e-mail: case_at_scripps.edu Dept. of Molecular Biology, TPC15 | fax: +1-858-784-8896 The Scripps Research Institute | phone: +1-858-784-9768 10550 N. Torrey Pines Rd. | home page: La Jolla CA 92037 USA | http://www.scripps.edu/case ==================================================================