Rule of thumb · ComputersNº 81 / 167

The ×100 latency ladder

RAM ~100 ns, SSD ~100 µs, spinning disk ~10 ms — each rung roughly 100× slower.

Why it works

Six orders of magnitude between memory and disk is why caching dominates systems design. One disk seek costs ~100,000 RAM reads.

When it fails

The rungs have moved. NVMe is nearer 20 µs than 100, and Optane closed the gap to RAM far further than the ladder allows. Queue depth matters more than the medium in many workloads: a spinning disk streaming sequentially can beat an SSD serving random 4 K reads.

More in Computers

← Previous3-2-1 backups Next →Every nine costs 10×

How slow is disk compared to RAM?

RAM ~100 ns, SSD ~100 µs, spinning disk ~10 ms — each rung roughly 100× slower. Six orders of magnitude between memory and disk is why caching dominates systems design.

Browse all 167 rules of thumb →