TI vs. Xilinx vs. picoChip

10/8/2007 5:01 PM EDT

In the fight for the high-performance DSP market there appear to be three main contenders. High-performance DSP processors are the entrenched incumbents, benefiting from a small army of engineers with DSP programming experience and huge amounts of existing code. FPGAs have in recent years emerged as a major challenger as high-end application requirements have outstripped single-core DSP performance. And newest to the party are a wave of massively parallel processor startups mainly targeting wireless basestations and emerging video applications.

Picking a winner is tough. Partly because, market forces aside, there's very little comparable performance data. That's why I was intrigued to see BDTI's recently released results for the massively parallel PC102 chip on its OFDM benchmark. This is the first benchmark I've seen with results for all three contenders. The PC102 from startup picoChip goes head to head with TI's flagship C6455 and FPGAs from Altera and Xilinx. Not surprisingly, the PC102 and the FPGAs fare much better than the TI DSP. The PC102 and FPGAs can implement many more OFDM channels than the C6455, which can implement only one, and offer much lower cost in terms of channels-per-dollar. The FPGAs beat out the PC102 in number of OFDM channels and channels-per-dollar, but not by nearly as large a margin as both beat the DSP.

Of course, many questions remain. Chief among them is the question of ease of programming, something much more difficult to benchmark. (This is a particularly pressing issue for FPGAs, as I noted in last week's blog.) For a discussion of programmability and other issues, such as the effect of advancing process nodes, check out FPGAs vs. DSPs: A look at the unanswered questions.