For a fair number of years now I’ve been following the twice-annual Top500 list of supercomputers which, as its name implies, rates and sorts the top 500 non-classified, participating supercomputers in the world. The rating is done using the LINPACK benchmark which focuses on floating-point linear algebra tests. Given those constraints, I like the Top500 as something of a rough measure of the current state of the human race’s technology civilization by measuring its capacity for high performance computing.
Today the November 2011 list came out (they release a list every June and November) and the reasonably smooth exponential growth that has occurred since the list began in 1993 has continued. They have included a poster with all the pretty pictures showing the top five and development over time. The Japanese K computer (after an intermediate upgrade) retains top billing and breaches the 10 PFLOPS barrier (because of what evolution did to our hands, the powers of ten are what seem to interest and attract us). Total aggregate performance has gone from 58.7 PFLOPS to 74.2 PFLOPS over the last 6 months. Remember, 1 PFLOP means 1 000 000 000 000 000 floating point operations per second, which is well beyond regular human reckoning. Technologically, things to note are that the x86 Intel and AMD platforms continue to dominate and that the use of GPUs is increasing (going from 17 to 39 over the last six months).
From the performance projection, it seems like Nov 2018 or June 2019 is about when we’d expect to see a 1 EFLOPS (where, of course, 1 exa = 1000 peta) supercomputer based purely on historical trends. With exabytes of data regularly crossing the Internet even now, exa- is the new prefix to remember in computing (it means quintillion if you’re keeping score).