HF
| QM Package | Time (min) | Steps | per step | Total E | HOMO | LUMO |
|---|---|---|---|---|---|---|
| erkale | 810 | 90 | 9 | -644.67570139 | -0.353712 | 0.074269 |
B3LYP
| QM Package | Time (min) | Steps | per step | Total E | HOMO | LUMO |
|---|---|---|---|---|---|---|
| erkale | 933 | 58 | 16.1 | -686.49566820 | -0.260899 | -0.064457 |
ERKALE is a very interesting project as it shows how the existing ecosystem of open source QM and maths libraries can be exploited as part of a modern QM package. But is it fast or slow? Who knows - we'll have to measure some more programs first...
Feel free to play along by forking the project and checking in your own results.
Note to self: The ERKALE version tested was SVN r1013 (2013-10-21). I had to do an extra single point calculation afterwards to find out the energy of the HOMO/LUMO. This time was not included in the assessment (as it shouldn't really have been necessary).
Great start!
ReplyDeleteI had a look at ERKALE now and packaged it for Debian. What I noticed when rerunning your calculations is that it appears to use OpenMP parallelization when building by default, and also during runtime.
Did you run the tests with OMP_NUM_THREADS=1, this is not clear from the input files you provided.
I didn't set OMP_NUM_THREADS. There is an erkale executable, and an erkale-omp one. I used the former which ldd says is not linked against libgomp. Looking at the automagic compile+download script, it seems that it builds both a parallel and serial version.
ReplyDeleteThis comment has been removed by the author.
ReplyDeleteI have now posted results for PSI4, MPQC and NWChem using the default Debian packages (and a backport for PSI4), as well as (newly packaged) Erkale HF results for reference here:
ReplyDeletehttps://github.com/mbanck/qmspeedtest