A growing number of products are incorporating computer vision capabilities. This, in turn, has led to rapid growth in the number of processors being offered for vision applications. Selecting the best processor (whether a chip for use in a system design, or an IP core for use in an SoC) is challenging, for several reasons.
First, these processors use very diverse architecture approaches, which makes it tough to compare them. Second, because vision applications and algorithms are also quite diverse, there is no single set of benchmarks that yields meaningful results for most applications. Third, processor offerings are emerging rapidly. And fourth, due to intense competition, processor suppliers often publish very little information on their products.
A client recently contacted BDTI for assistance in selecting a vision processor for a next-generation design. The VP of engineering faced a classic dilemma: It was critical to choose the processor that would give the best combination of high performance, low power, low cost, good programmability and low risk. But it was equally critical to move quickly, due to an aggressive development schedule and plenty of other tasks to be done.
Fortunately, BDTI's ongoing tracking of processor suppliers gave it the ability to move quickly and with confidence. Step one was to find all of the candidates; the client had identified some companies, but had overlooked other strong options. In addition, BDTI's trusted relationships with suppliers enabled BDTI to connect the client with one vendor whose unannounced processor under development proved to be an interesting contender.
Next came comparison of the candidates' offerings. The schedule was so tight that BDTI's preferred evaluation approach—collecting software development tool suites, simulators and emulators, writing code and implementing benchmarks—simply wasn't feasible. There wasn't even time for the vendors to do this work themselves. Instead, BDTI looked at benchmarking work the vendors had already done, and selected a set of algorithms that enabled extensive re-use of existing optimized code to create a set of benchmarks relevant to the client's application.
Leveraging existing code drastically reduced the work required by the processor vendors, but didn't completely eliminate it. Once the data started flowing in, BDTI's engineers noticed discrepancies that required further analysis. On what process node was the speed, power consumption, silicon area and other information gathered, for example, and at what lot-to-lot variation "corner" of that process node? Did the power consumption include the entire on-chip clock distribution network, the external memory controller, the system bus interface, and other peripheral circuits? These and other questions needed to be addressed with each candidate before a meaningful comparison between candidates was possible.
Using this combination of leveraging existing code and independently analyzing data to ensure apples-to-apples assessments, BDTI was able to provide a comprehensive, detailed comparison of the processor options in less than 90 days. The client declared the project an unqualified success, saying "We are very happy with the way you handled the project and results we achieved. This gives us great ideas on how to move forward."
As this case study shows, if you're selecting a processor, core or SoC for a vision processing application, BDTI is your one-stop shop for introductions, advice and evaluations. To leverage BDTI's unique and comprehensive understanding of the embedded vision processor market, and world-class benchmarking skills, contact Jeremy Giddings at +1 (925) 954-1411 or giddings@BDTI.com.