AMBER Archive (2005)

Subject: Re: AMBER: PMEMD limitation on NFFT

From: Chng Choon-Peng (choonpeng_at_cb.k.u-tokyo.ac.jp)
Date: Fri Nov 04 2005 - 21:44:58 CST


Dear Dr Duke,

     Thanks so much for the reply.
It is great to hear from you.

Actually I have increased gridhi from 256 to 512 already.
The line is
   parameter (gridlo = 6, gridhi = 256)
in:
/src/pmemd/src/mdin_ewald_dat.f90

I looked among the source codes but couldn't find the reason why the
limit
was set at 256, so I doubled it.
As you mentioned, possibly it was just a matter of 'history' since
F77 uses
static arrays.

The recompiled PMEMD now runs w/o errors and my system seems
stable.
Unfortunately, we're using just Gigabit Ethernet for our dual-core
Pentium4 cluster.

I'm interested in doing steered MD on a not-so-globular protein, so
employing
an exceptionally long unit cell.
Because of this, axis optimization is useful.

Thanks again for the prompt reply.

regards,
Choon-Peng

On 2005/11/05, at 12:29, Robert Duke wrote:

> Choon-Peng -
> Since everything is dynamically allocated in pmemd, I am not aware
> of any problems you will hit; just increase the limit and see what
> happens. The limits are values I picked up from the sander code
> and did not change; I would guess that the origin is in the volume
> of fft data you would have if you had a 256**3 grid space, and
> machine memories have gotten larger, and you are not pressing the
> limit in 3 dimensions. Sounds like an unusual unit cell though,
> and a big problem. Multiple processors would be good, and a fast
> interconnect to handle the volume of distributed fft data. Let me
> know how it goes; I am always interested in pressing the limits on
> size and speed with this stuff.
> Regards - Bob Duke
>
> ----- Original Message ----- From: "Chng Choon-Peng"
> <choonpeng_at_cb.k.u-tokyo.ac.jp>
> To: <amber_at_scripps.edu>
> Sent: Friday, November 04, 2005 7:38 PM
> Subject: AMBER: PMEMD limitation on NFFT
>
>
>> Dear AMBER users,
>>
>> I'm using a very long rectangular solvent box for my simulations
>> and I wish to employ PMEMD (AMBER8) for production run as it'll be
>> more efficient than Sander for such a case.
>>
>> However, I came across this error:
>>
>> | ERROR: nfft1 must be in the range of 6 to 256!
>>
>> The box I'm using has dimensions:
>> Box X = 410.713 Box Y = 93.892 Box Z = 55.886
>> Alpha = 90.000 Beta = 90.000 Gamma = 90.000
>> NFFT1 = 432 NFFT2 = 96 NFFT3 = 60
>>
>> I wish to know if I can simply increase the max size in PMEMD code,
>> from 256 to 512?
>> (I know where to change it)
>>
>> Can anyone please advice?
>> Thank you very much.
>>
>>
>> regards,
>> Choon-Peng
>>
>> --------------------------------
>> Choon-Peng CHNG
>> Doctoral Student (Yr 1)
>> Department of Computational Biology
>> Graduate School of Frontier Science
>> Affiliated with
>> Lab. of Molecular Design
>> Room 107,
>> Institute of Molecular and Cellular Biosciences
>> 1-1-1, Yayoi, Bunkyo-ku, Tokyo 113-0032, Japan
>> The University of Tokyo
>> Tel: +81 03-5841-1472
>> http://www.iam.u-tokyo.ac.jp/MolDes/indexE.html
>> ---------------------------------
>>
>> ---------------------------------------------------------------------
>> --
>> The AMBER Mail Reflector
>> To post, send mail to amber_at_scripps.edu
>> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu
>
>
> ----------------------------------------------------------------------
> -
> The AMBER Mail Reflector
> To post, send mail to amber_at_scripps.edu
> To unsubscribe, send "unsubscribe amber" to majordomo_at_scripps.edu

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