Unified Memory
In a traditional desktop PC, the processor (CPU) and the graphics card (GPU) live in separate worlds with separate memory pools—system RAM for the CPU and VRAM for the GPU—forcing data to be copied back and forth between them. Unified Memory eliminates that wall by placing a single, shared pool of high-speed memory directly into the main chip package so both the CPU and GPU can read from the exact same space simultaneously.
For local AI, this architecture solves a major problem: standard consumer graphics cards top out at 24GB to 32GB of VRAM, making it expensive and noisy to run massive 70B or 120B parameter models on a PC. Unified Memory machines—such as Apple Silicon Macs (M-series Pro, Max, and Ultra), AMD Ryzen AI Max+ mini PCs, and NVIDIA DGX Spark—let the GPU tap into 64GB up to 512GB of shared memory inside a quiet, low-power box.
There are two practical trade-offs to keep in mind before choosing a Unified Memory system. First, because the memory chips are integrated right onto the processor package at the factory, you cannot upgrade the capacity after purchase. Second, Unified Memory generally has lower bandwidth than flagship discrete graphics cards, meaning it excels at fitting gigantic models into memory rather than generating tokens at maximum speed.
Looking for verified hardware rigs?
Check our real-time £/GB VRAM index to find systems optimized for local model inference.