Unified Memory Architecture
A hardware design in which the CPU, GPU, and neural engine share a single high-bandwidth memory pool, reducing data copying, lowering latency, and improving energy efficiency, though memory capacity cannot be upgraded afterward.
Core Facts
Timeline (last 90 days)
The unified memory architecture of Apple's M-series chips makes it practical to run language models in the 7B to 13B parameter range locally on a MacBook, with latency controllable within a few seconds.
On PC platforms, when VRAM is insufficient, part of the model can be offloaded to system memory, reducing speed but still allowing it to run; on Mac, when unified memory is insufficient, the model immediately reports an error and exits.
A unified memory pool allows the GPU to directly address all system memory without needing to frequently move data between host memory and video memory.
All Facts (4)
Apple Silicon's Neural Engine shares a Unified Memory Architecture with the CPU and GPU, eliminating the bottleneck of data transfer between different memory pools in traditional heterogeneous computing.
75%UnverifiedThe unified memory architecture of Apple's M-series chips makes it practical to run language models in the 7B to 13B parameter range locally on a MacBook, with latency controllable within a few seconds.
50%UnverifiedOn PC platforms, when VRAM is insufficient, part of the model can be offloaded to system memory, reducing speed but still allowing it to run; on Mac, when unified memory is insufficient, the model immediately reports an error and exits.
50%UnverifiedA unified memory pool allows the GPU to directly address all system memory without needing to frequently move data between host memory and video memory.
50%