|
Guest Opinion
July 2000
Final exam
EEMBC aims to help product-development teams select processors and compilers via a series of free, application-specific benchmarks.
Alan R Weiss, EEMBC
Certification Laboratories
Convergence devices have decidedly different computational requirements than PCs and workstations. Like PCs, these products—from portable devices to automotive telematics systems to network equipment—will sell in huge volumes. But generally, they have no software legacy that prescribes specific processors or software. Design teams are free to choose the best hardware/software combination for the task at hand.
EEMBC (the EDN Embedded Microprocessor Consortium, pronounced "embassy"), an organization funded by a coalition of companies including processor and software vendors, aims to help product-development teams select processors (including DSPs) and compilers via a series of free, application-specific benchmarks.
While the EEMBC benchmarks are organized according to application area, two features reinforce the notion that applications are converging in new and interesting ways.
First, EEMBC discloses benchmark scores individually, rather than as artificial aggregates. Product development teams can pick the specific benchmarks that best suit their target products.
Second, EEMBC allows its members to publish both "out of the box" scores (which result from running standard C-code versions of the algorithms) and "full-fury" scores (based on algorithms the members can tailor for optimal performance on their own processors). EEMBC Certification Laboratories, an independent lab dedicated to assuring fairness and honesty, certifies both types of scores. Comparing the two scores gives development teams a good idea of what they can expect from compilers.
To see how EEMBC might benefit a product team, consider a convergence application: in-vehicle telematics (sometimes called intelligent transportation systems). EEMBC's automotive/industrial benchmark suite contains algorithms that engine controllers commonly execute, such as pulse-width modulation and angle-to-time. But curiously, it also contains algorithms such as IDCT (inverse discrete cosine transform), FFT (fast fourier transform), FIR (finite impulse response), and others more closely associated with audio and video processing. Why? Because EEMBC realized that the automobile was about to undergo a significant convergence with digital consumer devices, such as PDAs and multimedia computers.
Imagine a team working on a telematics unit incorporating speech recognition, a DVD player, and a digital cellular phone. The team might use the IDCT benchmark to reveal the processor's suitability for MPEG-2 decoding, then add the FFT and FIR benchmarks, which are important in speech recognition. Next, the team might use the telecom suite's FFT and Viterbi-decoder benchmarks to see how well the chip handles cellular-phone functions, then look at the consumer Bezier benchmark, which indicates the processor's potential for drawing curves on a map display.
In this way, the product team can create a customized and meaningful aggregate "figure of merit." By factoring in other design constraints, the engineers can get a pretty good idea how a particular processor might perform.
Useful as they are, EEMBC's certified benchmarks don't solve all design problems. For instance, in the example above, power consumption and heat dissipation would be challenging. EEMBC is working on that, too, with a new power-consumption-measurement system.
Author information
Alan R Weiss is Chairman of EEMBC
Certification Laboratories
| 
