Not many of us could convince our bosses that we’re most efficient when working slowly. But then, we aren’t microprocessors. For decades, researchers have known that processors achieve peak energy efficiency when their transistors operate at very low voltages near the threshold between their on and off states. Unfortunately, the circuits also become unstable at those low voltages. Nevertheless, Intel is trying to apply that research to future products.
The technology is called near-threshold voltage (NTV) computing. In effect, it’s extreme underclocking. PC users are more familiar with overclocking—cranking up the processor’s clock frequency and/or voltage to squeeze out more performance. Although overclocking works, it consumes more power—a small price to pay when we’re hungry for speed.
However, the price isn’t always small. Big data centers are sometimes limited by the electrical grid’s local capacity, and their cooling often gobbles more power than the servers do. At the other end of the spectrum, mobile devices must save energy to lengthen battery life and to keep from becoming too hot to handle.
Today’s processors can’t operate much below 1.0V. Intel has an experimental x86 processor (Claremont) rumored to work at 0.6–0.7V. Although that difference may seem tiny, it’s actually huge. Intel says the chip remains fully active (not in sleep mode) while sipping less than 10 milliwatts—low enough to run on a small solar cell.
The sacrifice is raw performance. Clock speed drops dramatically, but power efficiency (performance per watt) is five times higher. When more performance is needed, boosting the voltage drives the processor to its full clock speed.
Future processors could run some or all CPU cores at NTV levels most of the time. Stabilizing the circuits requires fine-tuning, though. With its own fabs, Intel has a potential advantage over competitors that outsource manufacturing to independent foundries.