The continued use of a single linear trend line suggests top500.org are still in denial about the dramatic slowdown in the growth of supercomputer performance since 2013, dropping from a 300× improvement each decade to just 20× (equivalent to doubling every 28 months rather than every 15 months):
Follow
@kate Wow, that's staggering! It's interesting that the slowdown knee happened so late; if I had to guess at a single cause, it would be the end of Dennard scaling in cutting-edge processes, but that would put the knee around 2005 rather than 2010, and that's certainly not what the data shows.
