Thoughts on passing a petaflop

As many of you have heard, we’ve passed a petaflop in performance for FAH.  A petaflop is 1 million-billion floating point operations per second, i.e. 10^15 operations per second or 1,000,000,000,000,000.  If we gave everyone on the planet a calculator, they would still have to do push over 200,000 buttons a second to keep up.  It’s a big number, but what does it mean and what does it get us?

First and foremost, it means that we’ve gotten a great turn out and we really thank all that our donors have done to help us.  But what really matters to all of us is not the turn out, but the results.  What a petaflop gets us in terms of the results is the ability to use very accurate models and have decent enough statistics to actually say something.  Usually in computer simulation, one has to choose one’s "poison": either a good model but no statistics (i.e. just one or two simulations) or a very simplified model (which may not be accurate enough at all) but with good statistics.  Both have major downsides.  With a PFLOP, we get to have it both ways — accurate models and the ability to actually say something about it.

We have several new results that have been submitted for publication that I think will show this off, but the unfortunate aspect of peer review is that it takes quite a while to get the paper out.  Nevertheless, peer review is a critical part of science and once we’ve passed that stage for this most recent work, I’ll be set to discuss it more publicly — which I’m very, very excited to do.