The fact that universities splurge money on these proprietary tools rather than spending it on long-term improvements isn't a good thing, even without the pain

The fact that universities splurge money on these proprietary tools rather than spending it on long-term improvements isn't a good thing, even without the pain of actually supporting them. The typical differences between stuff using ifort/MKL and gfortran/OpenBLAS/BLIS are at least similar, and probably smaller, than the sorts of variation you see anyhow on HPC systems, even without the important large-scale optimizations.