LocalLM Fit

Apple silicon index

Mac memory is shared memory.

A Mac does not have a separate VRAM pool. The GPU, CPU, operating system, and your applications all draw from the same unified memory. These pages use that capacity honestly, show the bandwidth of each chip tier, and keep MLX commands separate from Ollama and llama.cpp examples.

M2

Mac mini M2 · 16 GB

16 GB unified memory · 100 GB/s bandwidth

Ollama · llama.cpp · MLX · verified 2026-11-01

M2

Mac mini M2 · 24 GB

24 GB unified memory · 100 GB/s bandwidth

Ollama · llama.cpp · MLX · verified 2026-11-01

M3

MacBook Air M3 · 16 GB

16 GB unified memory · 100 GB/s bandwidth

Ollama · llama.cpp · MLX · verified 2026-11-01

M3

MacBook Air M3 · 24 GB

24 GB unified memory · 100 GB/s bandwidth

Ollama · llama.cpp · MLX · verified 2026-11-01

M4

Mac mini M4 · 16 GB

16 GB unified memory · 120 GB/s bandwidth

Ollama · llama.cpp · MLX · verified 2026-11-01

M4

Mac mini M4 · 24 GB

24 GB unified memory · 120 GB/s bandwidth

Ollama · llama.cpp · MLX · verified 2026-11-01

M4

Mac mini M4 · 32 GB

32 GB unified memory · 120 GB/s bandwidth

Ollama · llama.cpp · MLX · verified 2026-11-01

Capacity is not the same as speed. A larger unified-memory configuration can hold a bigger model, while a higher-bandwidth chip can generate faster when both configurations fit. Read the calculation limits.