Stroll the halls of any computer engineering graduate school and you will doubtless encounter numerous students brimming with ideas for cranking up processor speeds. Until recently, graduates of these schools found a warm welcome in the PC processor market, which seemed to have an insatiable need for speed. Today, however, fewer and fewer PC buyers are willing to pay a premium for more speed—after all, who needs a 3 GHz processor to write a letter to Grandma? If the PC market no longer needs ever-faster processors, perhaps these computer engineering students should have majored in art history instead.
In reality, computer architecture grad students need not worry. While the lust for ever-faster PC processors may be waning, communications applications still have a nigh-limitless demand for computational horsepower. Communications applications are constrained by inherently limited resources, which creates strong economic incentives to use these resources efficiently. For example, there is a limited amount of radio spectrum, so squeezing twice as many users onto the same amount of spectrum is extremely valuable. Fortunately for the up-and-coming processor architect, this need for efficiency translates directly into a need for processing power. For example, more-efficient radio spectrum use can be obtained via more-sophisticated—and hence computationally intensive—modulation, coding, and compression techniques.
Alas, the signal-processing work at the heart of communications applications has little in common with typical PC applications. As a result, the standard bag of tricks employed to improve PC speeds—deep pipelines, complicated branch prediction schemes, and the like—are not appropriate for communications applications. Worse yet, communications applications demand enormous processing power on shoestring budgets. While PC processors command multi-hundred-dollar prices, communications applications typically impose low-double-digit budgets on the processor. And while PC processors often consume dozens of watts of power, most communications processors are stuck with power budgets of a mere fraction of a watt.
The good news for processor designers is that there is much challenging work still to be done. The bad news is that the new challenges are quite different from the old. At least processor designers don’t have to fall back on that interest in art history.
Add new comment