What is being measured
Throughput, not latency: total bytes hashed divided by elapsed wall-clock time, reported in megabytes or gigabytes per second. The benchmark hashes the same data buffer many times and divides by the clock, so the figure is an average, not a single-run number. Buffer size drives the result - small buffers stay resident in the CPU cache and look unrealistically fast, while large buffers eventually hit memory bandwidth.