ICODE-ADC/lammps
4
0
Fork 0
Code Issues Pull Requests Actions Packages Projects Releases Wiki Activity

more details in serial performance summary

Browse Source
This commit is contained in:
Axel Kohlmeyer
2025-05-27 18:07:48 -04:00
parent d6a0d3bacf
commit 5ff0f05933
1 changed files with 7 additions and 6 deletions
Download Patch File Download Diff File Expand all files Collapse all files

13
doc/src/Speed_measure.rst
View File

@ -154,12 +154,13 @@ and change several settings incrementally:
#. Use tight PPPM convergence (1.0e-6 instead of 1.0e-4): 161.9 katom-step/s
#. Use Ewald summation instead of PPPM (at 1.0e-4 convergence): 19.9 katom-step/s
The numbers show that gains from aggressive compiler optimizations are rather
small in LAMMPS, the data access optimizations in the OPENMP (and OPT) packages
are more visible. On the other side, using more accurate force field settings
causes, not unexpectedly, a significant slowdown. Finally, using regular Ewald
summation causes a massive slowdown due to the bad algorithmic scaling with
system size.
The numbers show that gains from aggressive compiler optimizations are
rather small in LAMMPS, the data access optimizations in the OPENMP (and
OPT) packages are more prominent. On the other side, using more
accurate force field settings causes, not unexpectedly, a significant
slowdown (to about half the speed). Finally, using regular Ewald
summation causes a massive slowdown due to the bad algorithmic scaling
with system size.
Examples comparing parallel performance
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^