Unified memory
Unified memory is a design in which the CPU and GPU share one pool of memory instead of each having its own. On Apple Silicon Macs and similar chips, it lets a local AI model use most of the system RAM as GPU memory.
Also called: Apple Silicon memory, unified memory architecture, UMA
By Divya Sharma · Checked Nov 15, 2026 · How we review
On a typical PC, the GPU has its own VRAM, separate from the system RAM the CPU uses. A model has to fit in VRAM to run fast, and consumer graphics cards carry far less VRAM than a computer has RAM. Unified memory removes the split: the CPU, GPU and other processors on the chip read the same memory, with no copying between pools.
Why it matters for local AI
Apple introduced unified memory to the Mac with the M1 chip in 2020. A Mac with a large amount of unified memory can load a model that would otherwise need several GPUs, which made Macs a popular choice for running large local LLMs. Apple's MLX framework is written for this design, and llama.cpp, Ollama and LM Studio all run on it.
Limits
Capacity is not speed. Unified memory usually has less bandwidth than the VRAM on a high-end discrete GPU, and generation speed for large models is mostly limited by bandwidth, so tokens per second can be lower even when the model fits. The GPU cores are also slower than a data-centre card, which shows in long prompts. The operating system keeps part of the memory for itself, so not all of it is available to the model. The memory is fixed at purchase and cannot be upgraded.
Versus CPU offloading
CPU offloading is a workaround on split-memory machines: layers that do not fit in VRAM are kept in RAM and run more slowly. Unified memory avoids that split by design.
Related terms
- VRAM
- VRAM is the memory on a graphics card. For AI work it holds the model weights, the KV cache and working data, and whether a model fits in VRAM usually decides which GPU you need to run it.
- Local LLM
- A local LLM is a language model that runs on your own computer, phone or server instead of a provider's cloud API, so prompts and data stay on your hardware and it works without an internet connection.
- CPU offloading
- CPU offloading runs a model that does not fit in GPU memory by keeping some of its layers in system RAM, or on disk, and using the GPU for the rest. The model runs, but more slowly.
- GPU
- A GPU (graphics processing unit) is a processor with many small cores built to run many calculations in parallel. Its speed at matrix maths makes it the standard hardware for training and running AI models.
- Tokens per second
- Tokens per second (tok/s) measures how fast a language model produces output. It is quoted either per user, the speed one reply streams at, or per system, the total output across all concurrent requests.