X
Innovation

Intel and the correct use of power

One of the major themes at this year's IDF in San Francisco is the use and abuse of power. As Intel delicately drip-feeds information about its chips into the world, the picture builds up: these new devices are far cleverer about managing themselves, watts and all, than any before them, and that cleverness extends into and out of the chips themselves.
Written by Rupert Goodwins, Contributor

One of the major themes at this year's IDF in San Francisco is the use and abuse of power. As Intel delicately drip-feeds information about its chips into the world, the picture builds up: these new devices are far cleverer about managing themselves, watts and all, than any before them, and that cleverness extends into and out of the chips themselves.

Take Clarksfield, the Core i7 Mobile processor. Four cores running at 2GHz, taking about ten watts apiece, and if you're running software that uses all four cores equally then that's how it's going to be. Use fewer cores, and giant transistors inside the chip forcibly disconnect the idlers from the power bus. Run software that wants a single thread but uses it a lot, and the connected core is given more volts and a faster clock frequency until it's using a great deal more than the ten watt standard allocation.

That sounds simple enough, but there are lots of subtleties. The operating system may want the hardware to be running in a general low power mode to prolong battery life, so it may not be appropriate for the chip to aggressively maximise performance like that. Or a certain pattern of core usage may violate temperature rules: running too hot reduces chip life by years. So Clarksfield has a complex rules-based power control system that takes all of the factors into account, works out the best way to configure the chip for each combination, and constantly monitors internal and external information to keep things running as best it can. Incidentally, this means that if your computer has poor or faulty cooling, your chip will throttle back to save its life at the expense of performance – which is why Intel is also testing fluid dynamic models of case cooling.

Look inside Mooresfield, the next generation handheld chip, and you'll find the same ideas but cast in a very different way. A lot of the power in a smartphone or mobile internet device is used outside the chip itself, in running the camera or card interfaces, in processing audio or outputting video, and there are lots of variations depending on how the device is being used at any one time. Conversely, you get a lot of benefit if you can do things very fast for a decent slurp of power, then shut down to the lowest level possible until the next time. In such cases, if you run the processor hard for a bit then idle it, the memory and other buses connected to it also stop using power when the main chip is idle – more benefits. But again, you can't get too hot during your peaks of use.

By having a smart power controller – in this case, outside the chip in the support circuitry -- that understands the various usage modes of the phone and the conditions in which it finds itself, Intel configures and controls what it calls power islands, combinations of state that are best suited for particular times and tasks. That controller takes information not only from the operating system but from the various components within the phone, so it knows how to divide up tasks that make best use of what's available.

These ideas have much in common. Hardware, software and user work together to decide on the best way to use power, and apply rules and policies that reflect the complexities of being an Internet-connected, multimedia-capable device which has thermal, battery and other practical limits on what it can do, when. The user's role in this is minimal – deciding at the highest level how to trade off performance for power consumption – and then only for battery powered devices. The smarts live in the operating system and in the controlling systems.

A particularly pleasant side-effect of this work is that because Intel sees Linux as a major part of its various strategies, it has to provide open information and open code that demonstrate exactly what it's up to. This information is hard-won, and by sharing it the company is helping the whole industry to make progress into smarter, more efficient power management and higher performance at a lower cost. Such openness, even if encouraged by circumstance rather than philosophy, is actually doing what intellectual property laws claim to do but so often fail at: making new information patently obvious to everyone, for the advancement of all.

A true and proper use of power.

Editorial standards