Jeff Bier’s Impulse Response—Better Late than Never?

After Odysseus left to fight in the Trojan War, the ever-patient Penelope waited nineteen years for his return. Delayed by a few snags on the way back, he did eventually make it home, albeit much later than expected.

This phenomenon is not unknown in world of microprocessors, in which new processors are often late—sometimes very late. This is a serious problem, particularly among high performance chips. Moreover, in addition to being late, these chips are often not up to spec when they arrive.

In many ways this is understandable. Developing any processor, and especially high-end processors, is a massive undertaking with more than ample opportunity for mishaps. But in their enthusiastic drive to promote new products, vendors often make overly aggressive and optimistic projections about availability and performance. This effort to strengthen their competitive position and win ever-important early customer support can ultimately hurt customers and backfire on vendors.

System developers often design systems in anticipation of a particular new processor becoming available at a certain time and with certain specs. If that doesn't happen, their plans can be severely disrupted. Entire companies have been doomed by delays in processor availability and chips that don't function as promised. Even delays of six months or less can be fatal in the cutthroat world of consumer products.

This problem, endemic in the processor world generally, is only amplified with DSPs. Because most DSP architectures are proprietary to a specific chip vendor, if a new processor isn't delivered on time it may be quite difficult for a system developer to switch to another processor. Even within a single vendor's product line, chips are often incompatible, not to mention greatly disparate in performance. This makes last minute processor changes almost impossible to accommodate.

Certainly processor vendors face a dilemma: they need to disclose new product information to their customers early, but such information is necessarily subject to change. At best, vendor projections are educated guesses. At worst, overly aggressive projections are misleading. Until vendors commit to giving their customers realistic projections—and updating these as their own expectations change— system designers must protect themselves. They must find a mast that will shelter them from the Siren songs that vendors sing.