llama.cpp Runs GGUF Models Natively on a Jailbroken PS5 GPU
Original source
Llama.cpp on a jailbroken PS5: run any GGUF model natively on the console
Hacker News →PS5LM ports llama.cpp’s Vulkan backend to the PS5’s RDNA 2 GPU, giving users with a jailbroken console (firmware 7.00–13.60) a local model server. It loads any GGUF file the console’s 11.4 GB of shared memory can hold, then exposes OpenAI’s /v1/chat/completions and Anthropic’s /v1/messages endpoints on port 8081, so a laptop, phone, or coding tool like Claude Code can point at the console and get responses with no PC, cloud, or account in the loop. A built-in planner reads each model’s metadata to assign the longest context and best cache that fit available memory, and the app can pull models straight from Hugging Face to internal, M.2, or USB storage.
Beyond plain chat, PS5LM also runs decision models (d1, Clef, OpenJev) through TypeSafe’s /v1/systemone API, returning typed choices and calibrated probabilities in a single forward pass rather than generated text. Image-capable models such as Gemma 4, MiniCPM-V, and d1 accept photos in both chat and decision requests. A TV-facing chat view streams every incoming request so the console doubles as a status screen, and a Console page charts temperature, power, GPU, and CPU load over two-minute or one-hour windows.
The project is C++ on ps5-homebrew-ui with ggml-vulkan compiled against PS5_Vulkan’s RADV driver, built on Linux. It is released under GPL-3.0-or-later, credits llama.cpp (MIT) and several homebrew contributors, and carries no Sony SDK files, keys, or firmware in the repository.
Read the full article
Continue reading at Hacker News →This is an AI-generated summary. Read the original for the full story.