53ON: Open-source OpenAI-compatible AI server for AMD Ryzen AI Max mini PCs
A developer has released 53ON, an MIT-licensed AI server built for AMD Ryzen AI Max+ 395 mini PCs with 128 GB of unified memory. It runs as a dependency-free Node.js process on top of llama.cpp and supports both OpenAI and Anthropic API formats. The software includes a control panel for loading models, issuing per-client API keys with token quotas, and forwarding requests to cloud providers when local models fall short. The author also publishes a custom llama.cpp build whose KV cache allocates memory as tokens arrive, not all at load time, which reduces initial memory use substantially on Vulkan. So far it has been tested only on Strix Halo hardware with Vulkan (RADV) on Ubuntu 24.04, and the author invites feedback and bug reports.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in