[CP2K-user] HFX runs much slower with OpenMP compilation

[Nicklas Österbacka]

Hi!

I am trying to figure out why my hybrid functional calculations run 
significantly slower on a recent 8.2 compilation when compared to an older 
6.1 compilation available on our cluster, even though GGA calculations run 
significantly faster on the newer one.

To rule out version differences, I compiled CP2K 7.1 with and without 
OpenMP support using otherwise identical arch files and ran the same 
PBE0-TC-LR+ADMM calculation with both. The psmp compilation spends ~33 % 
longer in the integrate_four_center_bin subroutine.

I have included the in- and output from CP2K as well as the arch files used 
for compilation in the attached .zip archive. I explicitly set 
OMP_NUM_THREADS to 1 for the psmp calculation.

Very thankful for any help or pointers!

All the best,
Nicklas
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210624/e4586c37/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: cp2k-bench-files.zip
Type: application/x-zip
Size: 48064 bytes
Desc: not available
URL: <https://lists.cp2k.org/archives/cp2k-user/attachments/20210624/e4586c37/attachment.bin>