Mac Mini M4 / M4 Pro

Description: Compact AI workstation with Unified Memory for local LLM inference

Website: https://www.apple.com/mac-mini/

The Mac Mini M4 is an excellent choice for local AI inference. With Unified Memory Architecture and M4 chip, it offers strong performance for LLMs with low power consumption.

Technical Specifications

• Unified Memory: Up to 64GB shared memory for CPU and GPU (M4 Pro)
• Memory Bandwidth: 120 GB/s for fast token generation
• Metal Acceleration: GPU acceleration without complex driver installation
• Energy Efficiency: Significantly lower than NVIDIA GPU setups

Performance

• LLaMA 2/3 7B: ~12 tokens/second (32GB RAM)
• LLaMA 3.1 8B quantized: ~28 tokens/second
• Model size: Efficient up to ~10 billion parameters

Benefits

• From $599 for base model
• No VRAM limitation thanks to Unified Memory
• Perfect with Ollama and llama.cpp
• Quiet, cool operation