From the following figure, you can calculate what the throughput is based on AI Engine simulations. Eight inferences are ran in 532.8us, which translate to ~15,000 inferences per second.