Lammps performance with different compilers
This is just a rough comparison of the influence of compilers on the performance of lammps code on a VIC3.
OpenMPI 1.4 is installed and used during the compilation. A short REMD job using 26 CPUs was tested. (I think I don’t need to attach the job files here, because this comparison as you will see is really rough. The idea is just to get some brief knowledge of the performance influenced by the different compilers. It’s absolutely NOT scientific.)
Results:
intel_compiler_11.0 (6’54”) > gcc_4.3.3 (8’23”) > pathscale_3.2 (10’06”)
VIC3 is equipped with Xeon. So the results seem nature, where intel compiler outperformed than the others. Another worthwhile finding is the intel compiler can produce a code 30% faster than pathscale compiler, which reminds me I should preferably use it on HPC using intel CPUs. No fine tuning was made, but I think that will not change the ranking.
If the method you used to run this benchmark can be reproduced by us we’re more than happy to strive for better performance. In fact please contact our support and open a bug
Thanks
./Christopher
Hi, Christopher,
Thanks for the comment!
Please do the test with lammps in parallel mode, which is the way people often use this software.
I’m interested to know your result. Please let me know if you need more information.