This local-versus-remote latency gap is why production database teams explicitly pin MySQL and PostgreSQL buffer pools with numactl --membind, and why AMD's chiplet-based EPYC processors expose even more NUMA nodes per socket than older monolithic designs, making thread placement decisions that used to be optional now performance-critical. Cloud providers like AWS expose this same topology on bare-metal instances precisely because ignoring NUMA locality can cost real customers a measurable multiple in throughput.