Local LLM Hardware

Local LLM on Apple M4 Max

36GB VRAM · Unified LPDDR5X · 410 GB/s memory bandwidth · 40W TDP · released 2024

At a glance

VRAM36 GB
MemoryUnified LPDDR5X
Memory bandwidth410 GB/s
TDP40 W
Released2024
2026 price (used / new)~$2400
TierApple Silicon flagship
Best forbest mobile + desktop in one, 36GB unified memory

What models fit on Apple M4 Max?

Memory bandwidth is the bottleneck for LLM token generation. The Apple M4 Max's 410 GB/s bandwidth gives roughly the throughput below at Q4_K_M quantization. Numbers are conservative and assume a single request (batch=1), 2048-token context.

ModelQuantizationFits?Approx tokens/sec
Llama 3.2 1BQ4_K_MYes (0.8 GB)~74 tok/s
Llama 3.1 8BQ4_K_MYes (~5 GB)~25
Llama 3.1 8BQ8_0Yes (~8 GB)~18
Qwen 2.5 14BQ4_K_MYes (~9 GB)~14
Mistral Nemo 12BQ4_K_MYes (~8 GB)~16
Qwen 3 32BQ4_K_MYes (~20 GB)~7
DeepSeek R1 distilled 32BQ4_K_MYes (~20 GB)~7
Llama 3.1 70BQ4_K_MCPU offload only (~42 GB)~1
Llama 4 Scout 109B (MoE)Q4_K_MOffload only (~64 GB)~offload

Token estimates use a simple bandwidth heuristic (~tokens/sec ≈ bandwidth_GB/s × model_efficiency). Real numbers vary with model architecture, batch size, and KV-cache size. Treat as ballpark.

Build recommendations for Apple M4 Max

Best model to download first

Llama 3.1 8B for everyday chat and coding; Qwen 3 32B for top quality on 24GB. Start with ollama pull llama3.1:8b.

Recommended inference backend

For Apple Silicon, use MLX (best performance) or Ollama (simplest setup). Mullama works via llama.cpp but won't beat MLX on Apple Silicon.

Other GPUs to consider

Sources

[1] apple.com/macbook-pro · [2] mlx-lm benchmarks