> this original was a toy/PoC and this new paper is much better

This one hundred percent, the original has very few optimizations over a completely naive implementation. It uses MPI and huge pages, that's essentially it.