AMD Launches Lemonade: Open-Source Local AI Server with GPU+NPU Acceleration, 521 HN Points

AMD / Hacker News·high signal

AMD's Lemonade is an Apache 2.0 open-source local AI platform providing OpenAI-compatible APIs for text, image, vision, and speech across NVIDIA GPUs, AMD Ryzen AI NPUs, and CPU backends. HN users on Strix Halo report 35-60 tok/s on large models like Qwen3.5 122B, with Lemonade beating Ollama by ~30 seconds on identical tasks. The 2MB C++ backend auto-configures for hardware and bundles llama.cpp, Stable Diffusion, and Whisper — though HN commenters flag NPU underperformance and ROCm stability issues as real limitations.