AMD Lemonade SDK 10.7 adds NVIDIA CUDA support
AMD’s Lemonade SDK has received a new update with support for NVIDIA CUDA GPUs. This is an interesting change for a local AI project backed by AMD, since Lemonade was primarily built around Ryzen AI NPUs, Radeon GPUs and x86 CPUs.
Lemonade is a local AI server that offers API compatibility with OpenAI, Anthropic and Ollama. According to AMD, it is designed to automatically configure optimized CPU, GPU and NPU backends for text, image and audio inference on a user’s system. It is available for Windows, Linux, macOS and Docker.
CUDA support added for Windows and Linux
With Lemonade 10.7.0, the project adds llama.cpp CUDA support on Windows and Linux. The release also adds stable-diffusion.cpp CUDA support on Linux and stable-diffusion.cpp Vulkan support on Windows and Linux. The project says this improves cross-platform compatibility across more GPU vendors.
Source: Lemonade (Github)
The changelog confirms that NVIDIA GPU support was added through a new “llamacpp:cuda” backend. The same release also includes fixes for NVIDIA Blackwell and PRIME detection on Linux, which suggests the CUDA path is already being adjusted for newer NVIDIA hardware.
The update also adds LMX-Omni model support for OpenAI API-compatible apps that can render multimedia output, including Open WebUI and AnythingLLM. Lemonade 10.7.0 also adds a native Prometheus endpoint for real-time statistics monitoring.
Another new feature is the “lemonade bench” command. The project says it provides LLM benchmarking across llama.cpp, FastFlowLM, vLLM and Ryzen AI software. This should make it easier to compare local AI performance across different backends on the same system.
Source: Lemonade (Github)
