NUMA stands for Non-Uniform Memory Architecture and is a relatively new concept in computer hardware. In order to save bus bandwidth, a NUMA configuration divides the responsibility for parts of the physical address space between several CPUs. A memory operation on a CPU's own part of memory is faster than a memory operation on a different CPU's part of memory which has to be marshalled one or more hops across the bus to the other CPU.

NUMA presents a challenge for adaptive memory management, since object placement on the heap placement becomes much more of an issue.