Hi,

Assumption is incorrect.

A Gamma-point calculation uses only real numbers. A k-point calculation
uses complex numbers. Hence your matrix memory footprint gets doubled. So
2x2 = 4, and then there is *some* overhead with k-points, i.e. the numerics
changes a bit since this is also done in complex numbers.
So all seems fine to me :)


Den fre. 16. aug. 2024 kl. 22.00 skrev Francisco Garcia <
[email protected]>:

> Dear users,
>
>
> I run a large job with perfect orbital distribution over processors
> (exactly 20 orbitals per processor).
>
>
> As you can see in the data appended below, it takes approximately 25
> seconds between SCF steps in the Gamma point run.
>
>
> For 2 k-points however, it takes approximately 125 seconds between SCF
> cycles (see below), which is 5 times the Gamma point time.
>
>
> This is confusing me because a 2 k-point job is technically 2 successive
> Gamma point jobs, so I was expecting the time between SCF steps to be
> around twice that of the Gamma point job. 5 times is way too high and I was
> wondering if there's anything I can do to circumvent this problem.
>
>
> Thanks!
>
>
> Gamma point
>
> Start of run             0.000
>
> -------------- end of scf step            36.845
>
> -------------- end of scf step            64.071
>
> -------------- end of scf step            89.467
>
> -------------- end of scf step           115.114
>
> -------------- end of scf step           140.442
>
> -------------- end of scf step           169.311
>
> -------------- end of scf step           197.407
>
> -------------- end of scf step           222.953
>
> -------------- end of scf step           248.311
>
> -------------- end of scf step           273.939
>
> -------------- end of scf step           299.205
>
> -------------- end of scf step           325.194
>
> -------------- end of scf step           350.430
>
> -------------- end of scf step           376.049
>
> -------------- end of scf step           401.498
>
> -------------- end of scf step           427.002
>
>
>
> 2 k-points
>
> Start of run             0.000
>
> -------------- end of scf step           137.140
>
> -------------- end of scf step           263.171
>
> -------------- end of scf step           388.183
>
> -------------- end of scf step           516.512
>
> -------------- end of scf step           641.262
>
> -------------- end of scf step           765.025
>
> -------------- end of scf step           891.643
>
> -------------- end of scf step          1016.512
>
> -------------- end of scf step          1140.165
>
> -------------- end of scf step          1264.910
>
> -------------- end of scf step          1388.616
>
> -------------- end of scf step          1512.917
>
> -------------- end of scf step          1637.145
>
> -------------- end of scf step          1761.197
>
> -------------- end of scf step          1885.839
>
> -------------- end of scf step          2009.728
>
> --
> SIESTA is supported by the Spanish Research Agency (AEI) and by the
> European H2020 MaX Centre of Excellence 
> (https://urldefense.com/v3/__http://www.max-centre.eu/__;!!D9dNQwwGXtA!Uv0cJa-gg72cN5j_k95nAXxIVfr8RET835tytviEZiMv4UMA7-bVjS5urrfjmViNK0jfQG-TTiKrtCvDfQ$
>  )
>


-- 
Kind regards Nick
-- 
SIESTA is supported by the Spanish Research Agency (AEI) and by the European 
H2020 MaX Centre of Excellence (http://www.max-centre.eu/)

Responder a