No speedup using Intel MKL libraries?

Faraz H fa... at
Wed Nov 1 18:26:03 UTC 2017

I built cp2k 4.1 using the toolchain script using default gcc 4.9. I then 
built another executable with the toolchain with gcc 4.9 and the MKL 
libraries ( Parallel Studio 2016 ). I do not see any speed difference 
between the two executable . 

I am running the H20-128.inp file. It takes around 7 minutes on a machine 
with 28 cpus (  Intel(R) Xeon(R) CPU E5-2680 v4 @ 2.40GHz ) .  Is this 
normal or should I be seeing some measurable speed improvement? Here is the 
output of ldd on the two executables:

gcc4.9 only:

[user at compute99 benchmark]$ ldd 
/Apps/users/cp2k-4.1_gcc49/cp2k-4.1/exe/cp2k.ssmp =>  (0x00007fffcddb0000) => 
(0x00002b51f394e000) => /usr/lib64/ (0x000000343a800000) => /usr/lib64/ (0x00000032ee400000) => /lib64/ (0x0000003656a00000) => /usr/lib64/ (0x00000032ed800000) => /lib64/ (0x000000343a400000) => /lib64/ (0x0000003656200000) => /lib64/ (0x0000003655e00000)
        /lib64/ (0x0000003655600000) => /lib64/ (0x0000003656600000)

gcc 4.9 with Intel MKL libraries ( statically linked ):

[user at compute99 benchmark]$ ldd /Apps/users/cp2k-4.1/exe/cp2k.ssmp =>  (0x00007fff261ff000) => /lib64/ (0x0000003656200000) => /usr/lib64/ (0x00000032ee400000) => /lib64/ (0x0000003655a00000) => /lib64/ (0x0000003656a00000) => /usr/lib64/ (0x000000343a800000) => /usr/lib64/ (0x00000032ed800000) => /lib64/ (0x000000343a400000) => /lib64/ (0x0000003655e00000)
        /lib64/ (0x0000003655600000) => /lib64/ (0x0000003656600000)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <>

More information about the CP2K-user mailing list