Definition
Plain language
A chip design where the main processor and the graphics chip share one common pool of memory instead of each having their own.
As stated in the literature
A memory architecture, prominent in Apple Silicon, in which CPU and GPU access a single shared memory pool rather than copying data across separate pools; relevant as a deployment target where standard GPU-centric serving stacks don't map cleanly.
Why it matters: It changes how data moves on a chip, which means serving software designed for separate GPU memory pools often doesn't run efficiently without adaptation.
For example, on an Apple Silicon chip the processor and graphics chip both read from the same memory pool, so data doesn't have to be copied back and forth between them.
Heard on the show
“New hardware — Apple Silicon's unified memory, custom accelerators.”Episode 027 — When AI Agents Build the Serving Stack: A Bet on Bespoke Infrastructure