Queries about input and output files, running specific calculations, etc.
Moderators: Global Moderator, Moderator
-
suen
- Newbie
- Posts: 9
- Joined: Thu Jun 04, 2015 3:59 pm
- License Nr.: 5-2109
#1
Post
by suen » Wed Jun 10, 2015 8:06 am
Dear all,
I have compiled the VASP.5.3.5 on our HPC with Intel Fortran , Intel MPI and Intel MKL on Intel Core. Then test the parallel efficiency, the result is:
Code: Select all
Node 1 2 3 4 5 6 7 8
CPU 8 16 24 32 40 48 56 64
Time(s) 3063.8 1689.1 1219.0 967.1 810.8 832.3 651.5 589.5
Speed-up 1 1.81 2.51 3.17 3.78 3.68 4.70 5.20
Efficiency 1 0.91 0.84 0.79 0.76 0.61 0.67 0.65
Speed-up=Time/3063.8
Efficiency= Speed-up/ Node
The simulation system contains 112 atoms.
You can find the efficiency is lower than 80% when the nodes reaches 4, and decrease to 65% on 8 nodes. I want to know whether it is normal or not ?
Any suggestion will be appreciated.
-
admin
- Administrator
- Posts: 2921
- Joined: Tue Aug 03, 2004 8:18 am
- License Nr.: 458
#2
Post
by admin » Wed Jun 10, 2015 11:57 am
The efficiency can be modified by parameters for regulation of parallelisation (NPAR, NCORE).
Make a variation of these parameters consistent with your system (number of orbitals, k-points...)
-
suen
- Newbie
- Posts: 9
- Joined: Thu Jun 04, 2015 3:59 pm
- License Nr.: 5-2109
#3
Post
by suen » Thu Jun 11, 2015 4:35 pm
admin wrote:The efficiency can be modified by parameters for regulation of parallelisation (NPAR, NCORE).
Make a variation of these parameters consistent with your system (number of orbitals, k-points...)
Thank you very much! it would be help, i will try it.
