Intel managed to surprise just about everyone at its recent Intel Developer Forum (IDF) in San Francisco by pulling out of the bag a completely new processor family aimed at ultra low-power applications such as wearable devices.
The chipmaker is still keeping many of its cards close to its chest, but the Quark processor is intended to power not just novelty items such as smartwatches and bracelets, but also sensors and other devices that may form part of the Internet of Things, the future vision of a hyper-connected world where potentially anything may be monitored and possibly even controlled remotely.
Quark is slated by Intel to be just one fifth of the size of the firm's latest Atom mobile processors and to use a tenth of the power, in order to make it suitable for wearable devices and sensors that will be expected to last for weeks or months on battery power alone.
This being Intel, Quark is being based on the old faithful x86 instruction architecture and fabricated with a 32nm process. But can Intel really deliver an x86 chip capable of driving wearable devices and sensors while consuming miniscule amounts of power? Intel has shown that it was capable of rising to the challenge in the past, but I have my doubts this time.
The x86 architecture is now over 30 years old and has changed almost beyond recognition since the first 8086 chip, which comprised about 29,000 transistors. In the intervening years, processors have grown massively in complexity, gaining extra functions that were once separate, such as the floating point co-processor, memory controller and memory caches.
Internally, the x86 core design has also changed radically along the way, with multiple instruction pipelines, branch prediction logic, and other complex circuitry that speeds up code execution, but which also means that the number of transistors required to make a processor has ballooned to over a billion in the latest Core family chips.
More transistors mean higher power consumption, and while Intel has done a good job of cutting the power consumed by each transistor through progressively smaller fabrication processes, you still wouldn't put a Haswell chip into a wearable device such as a smartwatch and expect it to have a battery life of more than a few minutes.
So the key question for Intel is this: how simple can you make an x86 processor without compromising performance or losing Intel's trump card, which is compatibility with the huge ecosystem of existing software and tools for the x86 architecture?
It's easy to see how Intel might design a chip from scratch to be compatible with the x86 instruction set using as few transistors as possible, but without all the branch prediction logic and other baggage, its performance would likely be sub-optimal.
The question of software compatibility is especially significant, since modern software often assumes that certain hardware features will be available. Operating systems often require a memory management unit (MMU), for example.
If Quark turns out not to be able to run existing embedded software for x86 chips, then it loses many of the advantages that would make it stand out from the crowd of already available embedded chip architectures.
And here we must consider ARM, which has already been in this space for many years and thus has the advantage of a great deal of software ecosystem support.
ARM's latest embedded chip design, the Cortex-M0+, is claimed to consume just nine micro amps (µA) per MHz of clock speed. Even at one tenth the power of an Atom chip, it looks like Intel's Quark will find that a tough challenge.
Geoengineering on the sea floor near glaciers would form a new ice shelf to prevent melting
Alterations in capillary blood flow can be caused by body position change
Curiosity rover is in 'normal mode' but not transmitting scientific data back to base
NatWest outage comes a day after Barclays' IT systems shut out customers and staff