Source Newsroom: IEEE Spectrum Magazine
Newswise — The extraordinary reliability of computers makes them useful for many tasks, especially ones that we error-prone humans would never be able to do well. But this wonderful quality comes at a price: energy. You see, to ensure that errors are so rare that programmers can safely assume they never happen, computers consume gobs of energy. But for many applications, that degree of perfection is superfluous. So computer designers should really build machines that can switch into an energy-saving, albeit somewhat error-prone, mode on demand. The required modifications wouldn’t be too difficult. What’s harder to figure out is how exactly a programmer could make use of such hardware. To explore that challenge, the authors’ research group at the University of Washington has developed a computer language they call EnerJ, which allows benign errors to occur every now and then while preventing potentially catastrophic ones. It’s their contribution to a new approach for boosting energy efficiency, called approximate computing.