Jailbroken PS5 used as local AI server via PS5LM

Developers have successfully utilized jailbroken PlayStation 5 consoles to function as local AI servers. By using the PS5LM project, which ports llama.cpp to the console, users can run large language models (LLMs) natively using the system's GPU.

An enhanced version, PS5LM Plus, utilizes the AMD RADV driver from the Mesa open-source project and the Vulkan computing interface. This optimization increases available memory from approximately 6GB to 11.4GB, allowing the console to run highly quantized models such as Qwen 3.8-27B. Due to the console's high memory bandwidth, the Qwen model can achieve generation speeds of 20.9 tokens per second. Smaller models, such as Llama 3.2 3B, have reached speeds of up to 124 tokens per second.

The system serves models across a local network via an API compatible with OpenAI and Anthropic standards, enabling other devices like laptops and phones to access the AI. While demonstrating the hardware's potential beyond gaming, the process requires a jailbroken console, which carries risks of hardware bans or PlayStation Network account suspensions.