EEMBC Contributors Advance Industry Benchmarks in Lockstep with Technology Trends

Thanks to the extraordinary efforts of these members, developers will gain the ability to better compare the strengths of processors and platforms using industry-standard benchmarks designed to establish fair and equitable metrics.

The need for industry-standard performance benchmarks is increasing as both the hardware and software used to develop systems grows in complexity. EEMBC, with the support of its members, is developing FPBench to measure the floating-point performance used in a wide variety of embedded applications, BrowsingBench to evaluate the plethora of smartphones and other connected devices, and AndEBench to compare Android-based platforms.

Five EEMBC members stand out for the significant impact they have had in the creation and advancement of these much-needed industry benchmark initiatives:

• Rajiv Adhikary of Analog Devices provided technical guidance that pushed the FPBench initiative to adopt a non-proprietary and transparent standard of “goodness” in order to achieve wide acceptance and usability. His extensive research and industry collaboration have brought a depth and thoroughness that strengthen the FPBench effort.

• Mansoor Chishtie of Texas Instruments Incorporated (TI) chaired the BrowsingBench working group where he dedicated himself to creating a credible and realistic benchmark for smartphone and tablet performance. He was also instrumental in defining the test framework that can measure real-world browsing performance non-intrusively and with repeatable accuracy. Under Mansoor’s leadership, this simple-to-use and highly effective BrowsingBench 1.0 provides the only cross-platform browsing benchmark in the industry.

• Mason Guy of Intel took an active role in the development of BrowsingBench, ensuring that the benchmark included workloads to expand the range of content delivered by tablets. Additionally, Mason provided critical analyses that became the foundation for working group decisions about the stability of BrowsingBench and defined system configuration functions for performing battery-life testing.

• Luther Johnson of Microchip Technology determined fair and reliable cross-platform accuracy requirements for FPBench critical to the evaluation of floating-point performance. Luther applied his understanding of libraries and compilers to evaluate and recommend math function libraries for software-only reference implementations, and tested and ported multiple libraries to make FPBench safe for 16-bit microcontrollers.

• Ronen Zohar of Intel provided detailed, hands-on analysis that helped establish the robustness and utility of AndEBench, allowing the benchmark to support multiple, disparate architectures with a single Android APK, including multiple, concurrent tool chains. With FPBench, he identified residual sources of single/double precision confusion, created architecturally neutral methods of establishing initial data values and preserved high degrees of accuracy in the reference output values.
Creating meaningful performance benchmarks is a task that crosses company and product lines and requires the pooling of expertise, time, and efforts of industry leaders. These individuals have contributed to the EEMBC effort over and above their responsibilities within their respective companies, and with the support of these companies, have contributed to the greater good of the industry.

“A successful consortium is made up of individuals who are dedicated to making a difference and moving the industry forward,” said Markus Levy, EEMBC president. “I would like to personally thank these individuals for their commitment and technical expertise that has been invaluable in defining many of the EEMBC benchmarks. Developers who need EEMBC benchmarks to determine which platforms to use for their designs are indebted to each of these volunteers for their outstanding and generous contributions.”