QuantumATK Forum

QuantumATK => General Questions and Answers => Topic started by: jchang on January 8, 2018, 00:44

Title: Parallelization of PDOS calculation
Post by: jchang on January 8, 2018, 00:44

Hi,

I have tested PDOS calculation using different number of cores.

It seems that assigning more cores does not help to reduce the running time.
I have tested 2, 4 and 40 cores.

In my calculation, I have converged potential calculation for a device.
Device is periodic in the device width direction (direction perpendicular to the transport direction).
So, I use several transverse modes in the calculation.
And # of transverse mode is 29.

I thought PDOS calculation might be parallelized in transverse modes, but it seems not because I don't see much speed up with more cores.
Can anyone tell me how parallelization of PDOS calculation is done?
If I understand this, I think I could assign cores more wisely.

Thanks.
jiwon

Title: Re: Parallelization of PDOS calculation
Post by: Ulrik G. Vej-Hansen on January 8, 2018, 14:34

PDOS calculations should parallelize over cores, but depending on the size of the problem and the hardware, it does not necessarily give a big boost. If you provide the exact scripts and log-files, I can try to comment a little more specifically.

Title: Re: Parallelization of PDOS calculation
Post by: jchang on January 10, 2018, 08:18

Thanks for your response.
I will check speed up with more different number of cores, and then will get back to you.

jiwon