NWChem Slower with Multithreading?

Viewed 68 times, With a total of 1 Posts

Clicked A Few Times

Threads 5
Posts 14

9:52:54 AM PDT - Thu, Aug 2nd 2012
I am running NWChem 6.1 on my personal workstation under Ubuntu 12.04. NWChem was built on my machine with openMPI 1.6 . I have an i7-2600K (4 cores, 8 MB cache) and 16 GB of RAM. I have noticed that when I set the number of processors to 8 NWChem jobs runs slower than when I set the number of processors to 4. For example for an EOM-CCSDT calculation on a water molecule I get the following: -np 8 : Wall time = 216 s -np 4 : Wall time = 156 s -np 1 : Wall time = 359 s On the other hand when I parallelize a custom C code for numerical analysis using OpenMP, the wall time is faster when np is set to 8. Intel claim that multithreading could sometimes lead to a decrease of up to 40% in performance, is this what is happening in this case?

Gets Around

Threads 1
Posts 124

10:53:02 AM PDT - Thu, Aug 2nd 2012
DouDou The parallelization framework used in NWChem reaches best performance when the number of processes is set equal to the number of physical cores (in some case, even better when using number of cores - 1) Edo

Who's here now Members 0 Guests 0 Bots/Crawler 1

AWC's: 2.5.10 MediaWiki - Stand Alone Forum Extension
Forum theme style by: AWC