Unified memory

Apple Silicon for local AI

CPU and GPU share one memory pool. Apple list prices are U.S. starting prices for the exact SKU when Apple published one; configure-to-order memory SKUs stay unknown instead of inventing an upcharge.

Catalogue review recorded · individual claims have separate datesStructured datasetSourceMethodology

These rows are memory configurations, not SKUs collapsed by product name. 16 GB and 128 GB of the same chip are not equivalent for local AI. Unified memory is shared with the OS — it is not VRAM. First-party tok/s are listed only for configurations that have a validated receipt — currently 0 Apple measurements.

MacBook Air 13-inch (M5, 16GB)
16 GB unified · 153 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
MacBook Air 13-inch (M4, 16GB, 8-core GPU)
16 GB unified · 120 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
MacBook Air 13-inch (M4, 24GB)
24 GB unified · 120 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
MacBook Air 13-inch (M4, 32GB)
32 GB unified · 120 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
Mac mini (M4, 16GB)
16 GB unified · 120 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
Mac mini (M4 Pro, 48GB)
48 GB unified · 273 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
Mac mini (M4 Pro, 64GB)
64 GB unified · 273 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
MacBook Pro 14-inch (M4 Pro, 24GB)
24 GB unified · 273 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
MacBook Pro 14-inch (M4 Pro, 48GB)
48 GB unified · 273 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
MacBook Pro 16-inch (M4 Max 32-core GPU, 36GB)
36 GB unified · 410 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
MacBook Pro 16-inch (M4 Max 40-core GPU, 64GB)
64 GB unified · 546 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
MacBook Pro 16-inch (M4 Max 40-core GPU, 128GB)
128 GB unified · 546 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
Mac Studio (M4 Max, 128GB)
128 GB unified · 546 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
Mac Studio (M5 Max, 128GB)
128 GB unified · 614 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
Mac Studio (M5 Ultra, 256GB)
256 GB unified · 1200 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats
Mac Studio (M5 Ultra, 512GB)
512 GB unified · 1200 GB/s
MLX: official docs · llama.cpp: community
Apple specifications and price caveats

Community support is not official support or a measurement on this exact machine. Memory available to a process varies with the operating system, other applications and runtime allocation limits; we do not assume a fixed percentage is usable.