Uniform memory access
Uniform Memory Access (UMA) generally stands for a memory architecture in multiprocessor systems . There is only one global memory that can be accessed uniformly by all processors. Ideally, each with the same bandwidth and latency , which is why such a system is also called a symmetrical multiprocessor system (SMP).
The concept is in contrast to NUMA , in which the access time to the memory depends on the location of the memory.
Since 2013, AMD has been using the designation hUMA (for heterogeneous Uniform Memory Access ) in relation to the cooperation between CPU and GPU in computation tasks , which can use the same system memory through cache coherence . A simpler programming model and less data copying between separate memory areas should have an advantageous effect.
The shared memory graphics popular in notebooks are referred to as Unified Memory Architecture . However, this involves a memory that has been combined for reasons of space or cost and - in contrast to Uniform Memory Access - not about the shared use of memory contents.
- Peter Bright. AMD's "heterogeneous Uniform Memory Access" coming this year in Kaveri , Ars Technica, April 30, 2013.