This thread is locked.Only browsing is available.
Top Page > Browsing
Memory Consumption
Date: 2019/05/01 14:17
Name: Reza

Dear Developers,
I tested my input file with two format of FRAC and Ang for my conventional scf and neb calculations.
Whenever I use FRAC, my memory consumption increases more than twice!

and for neb calculation, whenever I use "MD.NEB.Parallel.Number 1" on my small system, compiler prints all the output twice. Whenever I use "MD.NEB.Parallel.Number 2" everything is OK. It seems that something in the code prevents that "MD.NEB.Parallel.Number 1" to be used and insist for image paralleling at least with two images.

I tested above points on 4-core system without thread, and 16 core system which has 8 real core and 8 threads. I tested with intel cluster studio 2015 and 2019 compilers.

Maybe above points help for the main problem of this interesting fast code which is memory consumption.

Thanks
Reza
メンテ
Page: [1]

Re: Memory Consumption ( No.1 )
Date: 2019/05/01 14:56
Name: Reza

I am confused!! May be my compiler has a serious problem. I used AU instead of Ang (converted all Ang data to Bohr), and get faster and less-memory results!! Could you please check it.
Thanks
Reza
メンテ
Re: Memory Consumption ( No.2 )
Date: 2019/05/01 15:35
Name: Naoya Yamaguchi

Dear Reza,

Usually, OpenMX begin to calculate systems after converting the all units of coordinates to AU, and the conversion is just a mathematical process, so I wonder if you might pass different ones to OpenMX. For example, if one change a parameter for "Atoms.SpeciesAndCoordinates.Unit", one should also change "Atoms.UnitVectors.Unit".

Regards,
Naoya Yamaguchi
メンテ
Re: Memory Consumption ( No.3 )
Date: 2019/05/03 02:42
Name: Reza

Thanks, Problem solved.
メンテ

Page: [1]