TURBOMOLE Modules > Define

Comparing Total CPU-time for different density functionals

(1/2) > >>

prasanta13:
Hi there,
I am trying to calculate and tabulate the time it takes to perform single-point energy calculation of a molecule by different density functionals. The DFT methods I use are b2-plyp, b3lyp (with/without -D2 and -D3), M06-2X, PBE (with/without -D2, -D3 (and ABC) and -D4) with def2-QZVPPD.
However, I am finding different results, as B3LYP taking significantly longer time (cpu-time = ~1 day 4 hours) to perform than B3LYP-D3 (cpu-time = ~ 22 hours). I have no idea.
Again, I was doing three consecutive job runs using same control, coord etc etc in different directory using same method and basis.
The runtime was (all cpu runtime), 1. 63490 mins, 2. 85263 mins, 3. 87830 mins.
This is very confusing. Why there is so much time difference?
I have not changed anything in bash (parallel methods), neither I have used different commands to call ridft. Everytime I used nohup ridft > ridft.out | tail -f ridft.out
There is no such difference in energy also.

What is the problem, can anyone help?
Thanks in advance and best regards.

Arnim:
Hi,
did you check the number of SCF cycles?
Also, do you run the job on a local scratch disk? And are there mayber other jobs running on the machine?

Cheers,
Arnim

prasanta13:
Actually, same numbers of scf iterations were performed.
The computer was same with nothing but turbomole was running.
I didn't set up any scratch directory explicitly, neither local nor remote.


Thanks Arnim.

uwe:
Hello,

did you use the parallel SMP version? Then the only valid output for timing is the wall time - which is the 'real' time the job needs. The CPU time is accumulated over all running threads and depends on a lot of factors, from my experience the numbers are very often meaningless...

prasanta13:
So I did the meaningless thing by taking the cpu-time for parallel run... :(

Thanks for the help Uwe...
Cheers

Navigation

[0] Message Index

[#] Next page