Lemonade SDK boosts AMD LLM performance 20%

AMD’s focused optimizations in the Lemonade SDK demonstrate that hardware-specific tuning is essential for maximizing the potential of modern NPUs and unified memory architectures. Direct integration with the XDNA 2 NPU and iGPU allows Lemonade to bypass the bottlenecks of general-purpose backends like llama.cpp. Achieving 90 tokens per second on a mobile workstation for cutting-edge models like Qwen3-Coder-Next makes complex local agentic workflows genuinely viable. By offering a lightweight, OpenAI-compatible API that integrates with VS Code and other popular tools, AMD is aggressively building a local-first ecosystem to compete with NVIDIA's developer mindshare.