AMD Lemonade SDK 10.7 Adds NVIDIA CUDA Support for Local AI

AMD has updated its open-source Lemonade SDK to version 10.7, introducing native support for NVIDIA CUDA backends on Linux systems. Version 10.7 allows developers to execute local AI tasks on NVIDIA graphics cards while maintaining support for AMD ROCm setups. This expansion allows developers to utilize NVIDIA graphics cards for running large language models and generative AI tasks without relying solely on proprietary software stacks.

New version enables local AI inference on NVIDIA GPUs alongside AMD ROCm setups

The Lemonade SDK serves as a framework for building and deploying machine learning applications across different compute architectures. Version 10.7 integrates the stable-diffusion.cpp project with CUDA backend support, enabling image generation pipelines to leverage NVIDIA GPU acceleration. The update also adds cross-vendor Vulkan extensions that improve compatibility when running graphics workloads on non-AMD hardware.

Key technical additions include a new lemonade bench tool designed for end-to-end large language model performance comparison across different inference backends. The new lemonade bench tool provides a standardized way to compare performance across different AI backends like NVIDIA CUDA and AMD ROCm. Lemonade SDK 10.7 introduces native support for the LMX-Omni model to streamline local AI processing.

The update introduces native Prometheus monitoring endpoints that provide real-time telemetry data for production deployments. The release adds Prometheus-compatible monitoring endpoints that track key performance indicators like GPU usage and memory allocation. This update preserves full compatibility with older versions of the SDK and introduces new options to assist developers in testing their code.

The updated SDK continues to serve as an open-source framework designed for flexible local AI development across different hardware platforms. By supporting multiple GPU architectures, the update gives developers greater freedom in choosing their preferred hardware for AI tasks. The release reflects a broader trend in the software community toward flexible, cross-vendor AI development tools.

AMD has confirmed the release of Lemonade SDK version 10.7 with verified CUDA backend integration and Prometheus monitoring capabilities. The update provides concrete tools for developers to benchmark AI workloads across NVIDIA and AMD hardware within a single software framework.