Researchers already have uses for supercomputers capable of exaflops (1018 floating point operations per second) - such as accurate global climate simulation - but existing technology cannot yet support it and the current trend won't make it practical. To reach achieve the goal, systems will have to be far more efficient with energy, memory, error-handling, and more.

 

Read the article by Robert F. Service at ScienceMag.org: What It'll Take to Go Exascale