It resembles sequential access memory with relatively fast seeks and a cache (ok, multiple layers of caches). Reading sequential, predictable addresses gives yo

It resembles sequential access memory with relatively fast seeks and a cache (ok, multiple layers of caches). Reading sequential, predictable addresses gives you much more throughput than random access, and reading a value that was recently accessed (or adjacent to such) is much lower latency than something that was not. There's further wrinkles in multicore systems as well, because then accesses to memory recently written to by another core can be slower again.