Deep breath, Todd, deep breath
Firstly, when HPL was designed, it wasn't pointless. It was pretty much the only portable benchmark that tested the capability of contemporary supercomputers to run an important class of problems.
Secondly, it's still not pointless, unless CPU speed has suddenly become irrelevent.
Thirdly, it's an excellent sanity check when throwing a cluster together:
- HPL efficiency a bit low? Hmm, let's run some interconnect tests
- Good check for excessive single-bit errors? Run HPL cluster-wide, as big as will fit on each system and see what gets reported.
Fourthly, EVERYONE knows the list is primarily used for business/marketing/dick-waving purposes. So? "Is vendor X's response to our RFQ any good?" "Yes, but they haven't built a machine on this scale since 2009." or "Yes, and I can see from the Top500 that they've got many machines in our sector at the scale we're looking." or.....
Fifthly, Jack Dongarra (et al) are perfectly aware that HPL isn't the be-all and end-all of benchmarking. That's why HPCC exists (and JD is the chair of its pretty stellar committee). Oh yes, and HPL is part of that improved benchmarking suite.
Sixthly, all benchmarks that aren't your application are - to a greater, or lesser degree - pointless. So let's not benchmark anything, eh?
Still, I'm sure your achievements dwarf those of Prof Dongarra....