Latency (top) and throughput (bottom) benchmarks of the inferencing accelerators