How many k-points in Brillouin zone you use? Is it molekel or surface (slab)
or bulk?
The calculation speed also depends on the speed of your interconnection band
between CPUs.
If data exchange between CPUs is fast (like in InfiniBand) then you can
expect some acceleration,
but if CPUs exchange da
In my structure, there are 118 atoms, and PAO.BasisSizes is DZP, the
calculate speed is too slow with 4 CPUs, I wonder whether adding 4 more CPUs
could save half of time?
Secondly, if my atom number doubled and other parameters unchanged, how many
CPUs is need to keep the speed fast?
在 2010年8月16日
It also depends on how you will use Siesta in parallel calculation.
There are two options:
1) parallel over K-points
2) parallel over orbitals
Check manual for syntax.
The first option works much better, but it is only useful when you need to
use many k-points.
The first option scales almost linear
DM.tolerance larger than default value ( 0.0001 ) should not be a good
setting.
Set your DM.tolerance as 0.0001 and try other things such as :
ElectronicTemperature 300 K # OR 600 K OR 900 K
DM.NumberPulay 6
DM.MixingWeight 0.1 # OR 0.01
PAO.BasisSize SZP
If yo
Dear all ,
In order to accelerate the speed of siesta and transiesta
calculation,
we are going to add the CPU number , I am wondering whether there is a
limit of CPUs or a saturation of calculation speed?
Thanks!
Holin Chen
16th August
