For something quantitative: the most time on general-purpose HPC systems is likely to be taken by materials science software written in Fortran. There are numbers somewhere for Archer, the UK "tier 1" system, for instance.
VASP is likely to be top. I've not seen much use of abinit. Gromacs, charmm, and siesta don't use Fortran. See also cp2k, elk, quantum espresso, nwchem, and dl_poly, amongst others, which do, and are common. Gaussian is unfortunately common, and I think is Fortran. That's on top of important Fortran codes in ocean and meteorological modelling, and engineering. https://www.archer.ac.uk/status/codes/ has a lot of "other", partly because they don't log what actually runs with whatever is in the Cray compute node kernel which would do a similar job to audit, for instance.
