[CP2K:414] Re: unique spline question

[Teodoro Laino]

HI Axel,

I had a look into the QMMM file you sent me..
First thing to notice: SPLINE_UNIQUE

UNIQUE_SPLINE .TRUE. (EPS 10^-12)
     0:   1>>>>>       4      1 force_field_control      44.230 0 Kb

UNIQUE_SPLINE .FALSE. (EPS default 10^-7)
     0:   1>>>>>       4      1 force_field_control     211.440 0 Kb

well unique_spline saves you at the end of the spline generation 4/5  
of the memory usage.. you may say that it's peanuts
since 200KB of memory is nothing.. I agree with you.. but the  
unique_spline is anyway much faster in generating only 1
spline and more important if you want a very high energy spline (EMAX  
several hundreads of hartree (like in the alchemical
free energy changes)) you can't do that because in that case the  
memory would really explode if you compute a spline for
each pair potential..

In your case the unique_spline just makes a little bit faster the  
initialization of the classical environment.

Cheers
Teo


On 29 Nov 2007, at 12:17, Axel wrote:

>
>
>
> On Nov 29, 6:05 am, Teodoro Laino <teodor... at gmail.com> wrote:
>> can you enable trace in &GLOBAL and send me the memory usage with/
>> without unique spline after the
>> generation of the second set of splines (first is 1-4 and second is
>> nonbonded)
>
> ok. will do so in a few hours when i get to the office. it is a bit
> tricky from
> home as i'm currently running a different test on the one machine that
> is
> easy to access...
>
> cheers,
>      axel.
>>
>> cheers
>> teo
>>
>> On 29 Nov 2007, at 12:03, Axel wrote:
>>
>>> i am not getting out of memory, i just don't save any. in fact the
>>> unique spline option costs a little more memory. ...and this applies
>>> to the virtual memory size _and_ resident set size as shown in top
>>> (when enabled, as VSIZE and RSS.).
> >