An EEMBC Benchmark


AudioMark provides a sophisticated porting layer that enables it to utilize available compute hardware in way that reflects real-world product architecture. For example, the main function can run on a host MCU while offloading the signal-processing functions to a DSP co-processor, and then send the results a neural net peripheral for inference. The "AudioMark" score itself reflects the total execution throughput of the platform, the higher the score, the faster the processing speed of the platform. The "AudioMark/MHz" score is the AudioMark score divided by the fasted execution clock used in the system, which characterizes the overall compute efficiency of the platform. For example, a high AudioMark score could be obtained by simply increasing the frequency, but even if the AudioMark score is higher than another platform at a lower frequency, that other platform might be a hardware accelerator that can perform more work per MHz, thus having a higher efficiency, and hence the usefulness of AudioMark/MHz. A platform may also run multiple instances of AudioMark to increase its overall performance. While parallel execution of each pipeline component is not allowed, multiple instances of the entire pipeline may be executed on multiple symmetric or asymmetric cores. Please refer to the Run Rules.

NOTE: We just launched, it will take some time for scores to start to appear on this table.

To compare scores, click the checkboxes in the table below, then click "Compare Scores".

{{ error }}
{{ first }} − {{ last }} of {{ total }}
Clear Sel.

Copyright © EEMBC