I'll admit it: I'm envious of TI's Gene Frantz. Not because of his technical contributions to DSP--well, maybe a little--but mostly because his status as Principal Fellow at TI, an elite group to which few are invited, has made him somewhat bulletproof at his company, a position we would all love to achieve.
Here's what I mean. As a preamble to his talk, "What to do with a billion transistors" at the Multicore Expo segment of ESC, Frantz was asked by seior management whether he was going to express his views or those of TI. "Mine," Frantz said, to which management replied "OK but can we at least look at your presentation to see if it is close to our view." Frants replied "Sure" but told the ESC audience that anyone who assumes that his talk represents the postion of TI "is wrong, It might be someday, but isn't now."
You've got to love this kind of independence.
According to Frantz success in multicore will hinge on a number of factors, including:
Moore's Law to determine the number of transistors on a processor and Amdahl's Law to determine the proper number of processor cores
Three vectors: greater performance, lower cost and reduced power consumption
Overcoming the bottleneck of processor to processor communications
Choosing a system design that solves a particular customer problem rather than trying to make the problem fit into a given solution
Frantz also reminded the group that multicore is not new. In 1990 TI developed the TMS7000 with 1,000 DSPs. Although they were all 1-bit DSPs the chip worked well in the early days of digital TV, processing each pixel and each line in a progressive scan or picture-in-picture application. Before too long, Frantz reminded everyone, compressed signal technology took hold and the processing tasks could be handled by a single DSP.