Processor designers know that a cycle-accurate simulator can be used to benchmark a processor that has not yet been fabricated. But many designers don’t realize that it’s also possible to benchmark an idea for a processor, a processor that may exist only in PowerPoint slides—and that there are good reasons for doing so.
As BDTI’s president Jeff Bier wrote in a recent column, BDTI has seen a number of cases in which a processor vendor used BDTI’s benchmarks on its fully-designed processor, only to discover that the processor’s performance wasn’t nearly as good as expected. One cause for such performance shortfalls is the lack of some esoteric feature required by the target application—perhaps a specialized data format, or a rounding mode, or some other capability. Unfortunately, there are any number of weaknesses—big and small—that can bring a processor to its knees.
The problem is that today’s applications are so complex that it’s difficult to identify all of the processor features needed for a given application without implementing the whole application—and that’s rarely possible in the course of developing a new processor. If the processor is an all-new design, or if it’s targeting an application area with which the vendor has little experience, then the risk of overlooking an important feature is much higher. And once the chip design is finished, it’s painful to go back and add missing features. That’s why it’s a useful exercise to benchmark the processor at the PowerPoint stage, when it’s still relatively easy to change the design.
Even before you have a simulator, you can study the benchmark reference code and think about how your processor would implement the required functionality. You can assess whether your processor has all the features you’re going to need for an efficient implementation, and use this exercise to help refine the architecture. If you implement the benchmark on paper—even in an approximate fashion—you can get a rough estimate of the processor’s performance. Then, when you start talking to potential customers about your fledgling processor, and they ask how it will perform in their application, you’ll have done the homework needed to provide them with credible answers.
The trick to all of this is to use benchmarks that capture all the important details of the application while distilling it into a form that is simpler to understand, and easier to implement (even if the implementation exists only as a thought experiment). A good benchmark is like a summary of an application’s most important characteristics; a sort of CliffsNotes for software.
Developing that kind of benchmark is a not easy. Most processor vendors don’t have the time to create good, accurate application-level benchmarks themselves, and even if they did, they wouldn’t have results for competitor processors to compare with. That’s where BDTI comes in.
BDTI has over 15 years of experience in developing benchmarks for embedded processing applications, and has invested a huge effort in ensuring that BDTI’s benchmarks capture the key performance characteristics of the applications they represent. As a result, BDTI’s benchmarks are relied on and trusted by embedded processor vendors and users worldwide.
Processor vendors can license BDTI’s benchmarks during the initial design stages to help refine their designs. Then, once the processor is finished and the benchmarks are fully implemented, BDTI can certify the final results. Certified benchmark results can be used in marketing materials to convince prospective customers that the processor has everything it needs to perform well in its target applications.
To learn more about BDTI’s benchmarks and licensing program, contact Jeremy Giddings at BDTI (giddings@bdti.com) or visit www.BDTI.com.
Add new comment