Someone raised the point that these are not like-for-like comparisons in terms of all of the program parameters (integration grid size, convergence criteria, etc, etc.). Pedro Silva investigated the effect of grid-size on the speed and results, and tried to normalise across GAMESS, Q-Chem and Firefly. However, our intention here is to give an idea of comparative speeds using the defaults as some sort of baseline, but also because it reflects typical use.
Unfortunately, I ran out of time to continue this project back in late Jan or so. The first victim of the time shortages were the Turbomole results. The folks at COSMOlogic came forward with an evaluation copy, but I simply ran out of time before having a chance to figure it out and use it. The "tmole" input files are already there though in the repo (from Michael Banck).
The others involved have been much more active. To see all of the results gathered so far, check out the Github pages of:
- Eric Berquist - Dalton, MPQC, NWChem, ORCA, Q-Chem
- Karol Langner - Dalton, GAMESS, Molpro, NWChem, ORCA
- Michael Banck - Erkale, MPQC, NWChem, PSI4
- Noel O'Boyle - Erkale, Firefly, NWChem
- Pedro Silva - GAMESS, Firefly, Q-Chem
Where to from here? Well, nowhere for me, at least for the immediate future. But these input files are out there now, and if you want to compare different packages, they make a good starting point. An interesting question is whether the relative speeds change once you start looking at larger systems. I don't know the answer to that...maybe you can try finding out.
