In a similar SPSC ring buffer setup I see something like 1.8ns per 16B entry without RDTSC (our design has a cached thread-local tsc that is updated periodically, so this is vaguely reasonable) or 7.6ns per entry with RDTSC (approximate RDTSC cost of 4.8ns). This is on a VM on some sort of Skylake CPU (maybe Skylake X).