TornadoVM 7 Enables GPU-Native Workloads from Java Applications (www.infoq.com)

TornadoVM has launched version 7.3, enhancing its open-source framework that enables GPU-native workloads for Java applications. This update allows Java developers to accelerate multi-core CPU and GPU processing without discarding existing Java Virtual Machines (JVMs). The framework now incorporates significant architectural improvements, using the Foreign Function & Memory (FFM) API rather than the Java Native Interface (JNI), simplifying access across multiple JDKs. A key feature is the optimized CUDA backend, which deepens integration with Nvidia technologies, enabling seamless calling of CUDA libraries and support for multi-GPU execution. The implications for the AI/ML community are notable, allowing developers to leverage GPU acceleration more effectively for data-parallel workloads, such as those typically seen in machine learning tasks. With enhanced library integrations (e.g., Apache Flink, Spring AI), TornadoVM provides a flexible programming environment that can adapt queries for GPU execution based on availability and suitability. Additionally, the introduction of jitLLM v2.0, a Java AI inference engine positioned as competition to existing solutions, enhances capabilities further. The TornadoVM team aims to produce a comprehensive JVM environment that taps into advanced computational libraries, facilitating efficient development and deployment of AI applications.