It's actually benchmarking 1000 repetitions of the RDTSC instruction running in parallel. My guess... On Skylake, multiple in-flight RDTSC instructions slow ea

It's actually benchmarking 1000 repetitions of the RDTSC instruction running in parallel. My guess... On Skylake, multiple in-flight RDTSC instructions slow each other down for some reason? Possibly because it's attempting to provide a strict monotonic guarantee, that no two RSTSC instructions will return the same timestamp. Intel's manual only claims monotonic, which theoretically allows for two RSTSC instructions to return the same timestamp.