Jeff Bier’s Impulse Response—DSP Vendors Need Multi-Core Tools Strategy

There’s been a lot of press lately about start-up companies offering multi-core DSP chips.  What’s less widely discussed is that large, established DSP chip vendors have been offering multi-core DSP chips for years.  These chips have been popular in “channelized” applications where workload partitioning is fairly straightforward.

But as multi-core DSPs move into a wider range of applications—and as the number of cores per chip grows—partitioning workloads among cores is becoming much harder.  This creates a challenge for established DSP processor vendors, whose chips have been the preferred solution for many applications in large part because they have been easy to use (at least, compared to the alternatives).  If the migration to multi-core designs makes DSPs significantly harder to use, DSP processor vendors will lose a key competitive advantage.

There are plenty of other companies poised to take advantage of any missteps. There's been a veritable frenzy of activity lately focused on creating novel strategies and tools for multi-core software development.  For example, several multi-core DSP chip start-ups have announced promising—though as yet unproven— programming approaches.  And in the so-called “high-performance computing” space, where multiprocessor servers tackle applications like financial market analysis, a variety of companies offer sophisticated tools to simplify application software development.

Mainstream DSP processor vendors risk being leapfrogged by other vendors who preemptively deploy effective multi-core development tools and strategies.  If that happens, DSPs may find themselves displaced from some of their traditional application strongholds. It wouldn't be the first time.

Back in the 1990s, DSP chip vendors ceded much of their military/aerospace market to high-performance CPUs.  Floating-point DSPs had been dominant in applications like radar, but when DSP vendors curtailed their investments in high-performance floating-point chips, system developers switched to general-purpose CPUs instead.  Arguably, that was a small market, and the DSP processor vendors could afford to let it go. But if they don't create a viable strategy to enable widespread use of their multi-core chips—soon—they may watch bigger markets go the same route.