Old smartphones repurposed as edge AI servers using Linux

Older high-end smartphones can be repurposed as efficient edge AI servers by installing Linux distributions such as postmarketOS. By utilizing a mainline kernel for ARM64 and running in headless mode, these devices can operate with minimal RAM consumption, often using only 180MB to 200MB.

This setup offers several technical advantages over single-board computers like the Raspberry Pi, including an integrated battery that acts as a built-in uninterruptible power supply (UPS) and very low power consumption, typically between 1 and 4 Watts. When running the Ollama framework, these devices can successfully host Small Language Models (SLMs). For example, Qwen 2.5 Coder 1.5B can generate approximately 12 to 15 tokens per second, while Llama 3.2 3B can achieve about 8 tokens per second.