To understand hUMA, we should first talk a bit about the UMA (Uniform Memory Access) and NUMA (Non-Unified Memory Access) architectures prevalent today. UMA refers to how the processing cores in a system view and access memory. In a UMA-capable architecture, all processing cores share a single memory address space. The introduction of GPU computing, however, created systems with Non-Uniform Memory Access, or NUMA. NUMA requires data to be managed across multiple memory heaps with different address spaces for CPUs and GPUs, which adds programming complexity due to frequent copies, synchronization, and address translation. Things come full circle with hUMA...



Link: HotHardware