Latency Numbers Every Programmer Should Know(opens in a new tab)
A table of typical costs, from an L1 cache reference (0.5 ns) and a main memory reference (100 ns) to a round trip within a datacenter (0.5 ms), a disk seek (10 ms) and a packet from California to the Netherlands and back (150 ms), credited to Jeff Dean and originally Peter Norvig and collected in Jonas Bonér's gist. The gist now also lists rough 2026 timings for local and hosted LLMs.