Miniaturization of chip components is only half the story. On the other side of the scale, you have the supercomputer: custom setups, built for power. In 2008, the IBM Roadrunner broke the one-petaflop barrier: one quadrillion operations per second [source: IBM]. (FLOPS stands for "floating-point operations per second," and it's the standard we use to talk about supercomputers used for scientific calculations, like the ones we're talking about in this article.)
Expressed in scientific notation, a petaflop is measured on a scale of 10^15 operations per second. An exaflop computer -- which experts predict will arrive by 2019 -- performs at 10^18, or a thousand times faster than the petaflop computers we're seeing now [source: HTNT]. For comparison's sake, as of June 2011, the 500 fastest supercomputers in the world combined would still have less than 60 petaflops of power. Continuing into the future, zettaflops improve on the same scale, giving us 10^21 operations per second by 2030, and then come the yottaflops, at 10^24 [source: TOP500].
But what do those numbers really mean? Well, for starters, it's believed that a complete simulation of the human brain will be possible by 2025, and within 10 years, zettaflop computers should be able to accurately predict the entire weather system two weeks in advance.