TornadoVM 7.3 Adds Multi-GPU Support for Java
The latest framework update streamlines AI inference by replacing the Java Native Interface with the Foreign Function and Memory API.
Updated on Oct. 21, 2026 in Artificial Intelligence
Live Poll
Do you plan to use Java-based tools for GPU-accelerated machine learning tasks?
The TornadoVM team has released version 7.3, a framework designed to run Java workloads on hardware accelerators. This update introduces the jitLLM v2.0 engine to streamline AI inference on JVM-based systems.
Why it matters
By providing an idiomatic JVM environment for hardware libraries like Nvidia's CUDA, this update aims to accelerate data-parallel workloads on GPUs. It enables Java developers to deploy high-performance computing tasks across multi-GPU setups without leaving their native ecosystem.
TornadoVM 7.3 replaces the Java Native Interface with the Foreign Function and Memory API and removes the experimental JVM Compiler Interface dependency. It supports multi-GPU execution alongside an optimized CUDA backend for Nvidia hardware.
The players
TornadoVM
An open-source framework that enables the execution of Java applications on heterogeneous hardware including GPUs and multi-core CPUs.
Beestream.ai
The commercial entity responsible for the professional support and commercialization of TornadoVM and the jitLLM engine.
The details
TornadoVM functions by compiling Java source code into execution streams for OpenCL and Apple Metal backends. The system gathers compiler metadata through a combination of reflection, ASM—a library for bytecode manipulation—and an Unsafe layer to manage low-level memory operations. This approach allows the framework to execute Java tasks directly on accelerators, bridging the gap between high-level JVM code and hardware-specific instructions.
Timeline
October 21, 2026: TornadoVM 7.3 and jitLLM v2.0 are released.
December 2025: InfoQ covers the TornadoVM project.
2022: InfoQ covers the TornadoVM project.
The Tech Race
TornadoVM 7.3 signals a shift toward modernizing how Java interacts with low-level system hardware by moving away from the legacy Java Native Interface. This update aligns the framework with current high-performance computing trends by integrating with major JVM ecosystems like Spring AI and Quarkus.
Java developers can now leverage multi-GPU support and new JDK compatibility to integrate AI inference engines directly into existing application workflows. The framework is designed for immediate integration with major enterprise stacks including Spring AI, Quarkus, and Flink.
The takeaway
The transition to the Foreign Function and Memory API reflects an effort to improve performance and stability for Java-based AI workloads. Developers should monitor the integration status of jitLLM v2.0 within their specific project pipelines to evaluate potential throughput gains.
Further reading
For more on the latest advancements in high-performance execution, explore Artificial Intelligence.
Live Poll
Do you plan to use Java-based tools for GPU-accelerated machine learning tasks?
