Whatever it's called or does, the only sure thing is that designers want it to go faster. The only problem is that performance is hard to predict, especially when you’re sorting through the mishmash of competitive pitches and claims.

Unlike the desktop market, the embedded world has never had the luxury of SPEC or similar commercial benchmark suites. Instead, there's a motley collection of Dhrystone, VAX MIPS, and native MIP, MAC, and FLOP claims that are so vague, incomparable, and unverifiable that they border on snake oil.

Enter EEMBC (EDN Embedded Microprocessor Benchmark Consortium, pronounced "embassy") with the holy grail of realistic and trustworthy benchmarks for embedded applications. Under the leadership of EDN’s Markus Levy, EEMBC is a consortium of more than two dozen micro suppliers, including all the big names. They've come up with benchmarks for five distinct markets, including the automotive/industrial (see Table 1), consumer, networking, office automation, and telecommunications.

EEMBC faces the inevitable temptation for suppliers to optimize (er, cheat) and faces it head on with an independent certification lab and extensive rules about what's kosher and what isn't, lest benchmarking end up being a programmers’ contest. There'll be both out-of-the-box and optimized scores. Out-of-the-box scores give you a way to compare different C compilers for a particular chip, whereas optimized scores represent the most likely outcome for a real-world design.

EEMBC makes their living off the suppliers, not users, so the benchmark results are posted publicly on their web site at www.eembc.org.

  PREVIOUS • NEXT