Dictionary entry
unified memory
also unified memory architecture or UMA
Definition of unified memory
-
: a memory architecture or programming model that lets a CPU and one or more accelerators access a shared address space or memory pool, reducing the need for applications to maintain and copy separate CPU and GPU versions of the same data
- The workstation's unified memory allowed the local language model to use a larger shared memory budget than its discrete-GPU counterpart offered as VRAM.
- The inference framework used unified memory to move data between the CPU and GPU without explicit copy commands in the application.
Origin & history
Unified-memory designs predate modern generative AI. NVIDIA announced Unified Memory as a CUDA 6 feature in November 2013, using a managed pool accessible by CPU and GPU, and Apple later made unified memory architecture a prominent feature of Apple silicon. In the 2020s, the term became increasingly visible in local-AI hardware because model weights and context can require large memory capacities.
Test yourself
Which of these is the meaning of unified memory?
Cite this entry
"Unified memory." AI Dictionary, Dadgogo, https://dadgogo.com/dictionary/unified-memory/. Accessed 9 Oct. 2026.