Intel's introduction of a 1-GHz processor marked a milestone in more ways than one: The new desktop processor is exactly twice the speed of Intel's best mobile processor. The mobile performance gap has been growing slowly over the past few years, but recent announcements have stretched it into a chasm. Despite Intel's fervent claims that "the performance vs. mobility compromise is now obsolete," there are no signs that the gap will close.
Intel's fastest mobile processor runs at just 500 MHz in battery mode. The same chip will run as fast as 650 MHz, but only if the system is plugged in to the wall. While that SpeedStep mode helps some users, it does nothing for truly mobile users on a train or airplane. And even in tethered mode, the premium notebook CPU falls five speed grades behind the best desktop processor.
An immutable physical law, namely p = cv2f, is mucking up the works. Power is rising because transistor count (which affects capacitance, c) and clock speed (f) are rising exponentially. Supply voltage (v) is not falling fast enough to compensate. Already as low as 1.35 V, the supply voltage is approaching a natural limit of about 1 V.
Intel's 1-GHz Pentium III, with 23 million transistors, burns 32 watts (max) at 1.7 V. To fit that chip into a notebook, Intel must cut the power to about 12 W by cranking down the voltage and the clock speed. Mobile performance suffers.
AMD's Athlon is no better; more than twice as hot as the Pentium III, the current Athlon can't fit into a notebook at any speed or voltage. The problem: Both companies design their processors for the desktop first and later try to squeeze them into mobile systems.
Desktop processors consume many amps to eke out the last few percentage points of performance. Many transistors don't really compute; they analyze and reorder code for optimal performance. In some cases, results are computed speculatively and then thrown away after a branch misprediction or other unexpected event.
The ideal mobile processor would rebalance the performance/power trade-off by avoiding code reordering, instruction analysis and speculative computation. That's the approach Transmeta has taken with Crusoe. Questions remain as to the performance and viability of Crusoe. But at least Transmeta tried to design an X86 processor for mobile computing.
To better serve the notebook PC arena, Intel and AMD must take a similar tack.
LINLEY GWENNAP IS THE FOUNDER AND PRINCIPAL ANALYST OF THE LINLEY GROUP (WWW.LINLEYGROUP.COM).