The Debichem pure blend has been packaging various computational chemistry packages over the last years. While their functionality varies, a lot of them implement basic quantum chemistry methods for use in material science, biochemistry and other fields. However, the maturity and efficiency of the codes vary as well and it is not always easy to differentiate which projects should be packaged and which can be dropped because their functionality is covered in other packages and they are not competitive performance-wise.
In 2013, Noel O'Boyle has initiated the qmspeedtest project, which aims to benchmark quantum chemistry packages. So far, the aspirin molecule gets optimized by two different quantum chemistry methods. The focus is on default settings (as usually used by non-export users) and single-core performance. Several researchers have since posted numbers for various (mostly proprietary) computational chemistry codes.
This talk will summarize the qmspeedtest results for the packages in Debian or under consideration by the Debichem team. Preliminary results show vastly varying runtimes between 20 minutes and 2,4 days for the same system for different codes. Possible further investigations include (i) expanding the benchmark to other molecules, (ii) assessing the parallelization speed-up (and the parallelization penalty for serial runs), and (iii) the influence of different linear algebra library packages (blas/lapack, Atlas, ELPA, mkl) or (iv) compilers (gcc, cblas, icc/ifc).