i like how the sources reveal external network technology, that the load is so much more than what can fit onto a single board, that, over generalizing here, it's actually probably a string of connected modules to help support the count of up to 1,000 processors.
and as far as the software, pretty sure, as underground news, many extremely small, but expensive private technology companies that are operating by the millions of dollars, even hundreds of millions of dollars, will try and program for breakthroughs like this - whether or not they yield anything or not, is a question of sometimes profit motive and direction of their research.
can't beat the DoD's ambitions. after all, they helped give us the M-16-A2.
also, I did research back in 2010/2011 for college work, and online sources at the time pointed to Pres. Obama wanting our super computers to reach so many flops/sec by 2018. i'm sure this is no coincidence for 2016, US tax dollars hard at work here to accelerate for faster and faster computers.
trillion instructions/sec, now imagine that... *-*