Seldom has a piece of aging hardware been repurposed so ingeniously. A developer named Leonardo Alegría has managed to run a sophisticated AI model on a PlayStation 4. The console, originally released in 2013, was never designed for such a demanding computational task. Yet through creative programming, it now functions as a fully offline AI chatbot.
The open-source project, called Garlic LLM, runs Alibaba's Qwen3.5-9B model entirely on the console's old graphics chip. This AI model contains nine billion parameters, making it remarkably powerful for such outdated hardware. In text conversations, it generates between thirty-one and thirty-nine words per second. For code generation, it reaches up to forty-three tokens per second.
What distinguishes this project is how it bypasses conventional software layers to communicate directly with the hardware. The PS4 shares just eight gigabytes of memory between its processor and graphics chip. To circumvent this limitation, the model was compressed to approximately 4.5 gigabytes using advanced techniques. The project's name itself derives from the GPU's direct memory pathway, known as "Garlic" memory.
The endeavor also includes voice conversation capabilities in seven languages, making it more than a mere text tool. However, running this software requires a jailbroken console, which voids the manufacturer's warranty. The developer has emphasized that Garlic LLM contains no exploit itself. Users who wish to experiment must assume the associated risks independently.
This achievement underscores a broader trend in the technology world. Resourceful programmers are increasingly finding ways to repurpose obsolete devices for contemporary applications. Had anyone predicted a decade ago that a gaming console could host billions of AI parameters, skepticism would have prevailed. Nevertheless, projects like Garlic LLM demonstrate that innovation often thrives within constraints rather than abundance.






