Mark Nadeski

To help evaluate the performance of our processors, we have created a new benchmark site that presents benchmark data from TI’s optimized libraries as well as independent benchmarking results. These benchmarks help illustrate the benefits of DSP cores when used to implement complex mathematical algorithms.

There are two main areas on the benchmarking site:

• Core benchmarks focus on the performance of a single core. Benchmark information for the C674x DSP, C66x DSP and the ARM® Cortex®-A15 cores are shown here.
• Device benchmarks illustrate performance at the device level. This shows how performance scales with the number of cores used and takes into account memory, data movement and cache effects.

There is also a TI DSP benchmarking application report that shows how to duplicate the core benchmarks shown. This document details the hardware, tools and libraries used in the benchmarking process and provides modified linker command files that can be used to replicate the results.

As we discussed in our third DSP Breaktime Video (see the 3:40 mark), comparing benchmarks between different types of processors can be a tricky thing, especially when it comes to FPGAs. To paraphrase Arnon: “You have to be very careful making comparisons based on isolated benchmarks, you need to compare apples to apples, not apples to oranges.“

How do TI's benchmarks compare against FPGAs and GPUS?

Our benchmarking page is just getting started. What other benchmarks would you like to see?