On February 6, 2026, the TornadoVM team from UNIMAN presented TornadoVM at Voxxed Days CERN, hosted at CERN in Switzerland.
In this session, we introduced TornadoVM as a state-of-the-art open-source framework that enables native GPU acceleration directly from Java, allowing developers to program heterogeneous hardware while the runtime transparently handles compilation, execution, and optimization. We also highlighted its real-world impact, including its adoption by the European Space Agency’s GAIA mission, where GPU acceleration has been integrated into large-scale data processing pipelines as part of our project.
A key highlight of the talk was GPULlama3.java, the first GPU-accelerated native Java library for LLM inference, integrated with LangChain4j and Quarkus. The session emphasized that Java is AI-ready, that GPU programming from within Java is now practical and accessible, and that a native Java platform for AI and general-purpose acceleration enhances both developer productivity and ecosystem integration.
