Benchmarking is a tricky business. There are benchmarks and there are benchmarks. I'm sure I'm not alone in my weariness with reading reports on yet another batch of benchmark results.
MADISON, Wis. — Benchmarking is a tricky business. There are benchmarks and there are benchmarks. I'm sure I'm not alone in my weariness with reading reports on yet another batch of benchmark results.
Among the more notorious examples is the recent news about Intel's Z2580 application processor, codenamed CloverTrail prior to launch, which outperformed competitors' processors in a benchmarking exercise. The report, issued by ABI Research in early June, concluded that Intel has succeeded in reducing significantly the power consumption of its smartphone application processor and now rivals equivalent processors based on the ARM architecture licensed from ARM.
Subsequent reporting and investigation, however, revealed that ABI's conclusions were derived from one outlying benchmark (done by AnTuTu). The market research firm neglected to compare results from a suite of benchmarks.
To be clear, in the electronics industry, there's no shortage of benchmarks. Benchmarking exercises are carried out by various outfits for just about every purpose -- from CPUs, GPUs, and DSPs, to FPGAs, and more.
"There are surprisingly very few benchmarks available that capture the performance of the whole phone," said Jeff Bier, president of Berkeley Design Technology, Inc. (BDTI). Dissecting the performance of technology on a component level is one thing. "But what about measuring technology in terms of what matters to consumers, such as battery life, application speed, and network performance for a variety of mobile phones and tablets?" Bier asked.
BDTI last week announced that it has teamed with Qualcomm to create a new user-experience rating for mobile devices such as smartphones and tablets.
For BDTI, a technology consulting firm known for developing its own benchmarking for processor cores and other technologies over the last 20 years, this ascent up the food chain to focus on system-level performance struck me as a radical departure. Bier, however, stressed that BDTI is uniquely qualified to meet the challenge precisely because of its decades of expertise in independent benchmarking, which has earned widespread trust in the technology industry. "We know what we're doing," said Bier.
Of course, I couldn't help but point out that BDTI is taking money from Qualcomm to develop this new benchmarking for the consumer experience of smartphones and tablets. How could it be "independent"?
Bier said, "Of course, no model is perfect." The new efforts are being funded by Qualcomm, but BDTI is independently designing benchmarks and experience ratings based on technical merit, he asserted.
"How we actually designed benchmarks will be transparent, and it can be viewed by network operators, OEMs, and savvy technology analysts" for their inspection. "In the interest of checks and balances, our policy is to let others see how we've done it," he added.
For those with legitimate business interests, the source code of BDTI's benchmarking will be available for a fee, said Bier. Journalists and industry analysts can see the results of the new benchmarks for free.
Bier also noted that BDTI is fully aware of many pitfalls and challenges associated with benchmarking efforts.
History suggests that all benchmarks are subject to manipulation. Even if they're developed within an industry consortium, there can be members who know ways to skew a benchmark in ways that stack the deck against competing technologies, pressure others, and get votes from their friends.