Java News Roundup: TornadoVM 4.0, Google ADK, and More

Java Ecosystem Evolution: Accelerating AI Workloads with TornadoVM 4.0

The demand for high-performance execution in the age of Large Language Models (LLMs) is forcing a paradigm shift in how Java applications handle computationally intensive tasks. Traditionally, Java developers faced significant engineering hurdles when attempting to move workloads from the CPU to the GPU. The release of TornadoVM 4.0.0 on February 4, 2026, represents a major leap forward in solving this bottleneck by enabling automatic acceleration across heterogeneous hardware.

What is TornadoVM?

TornadoVM is an open-source plug-in for OpenJDK and GraalVM designed to allow programmers to run Java programs on heterogeneous hardware without requiring them to write low-level kernel code. It automatically accelerates Java code across multi-core CPUs, GPUs, and FPGAs, reducing the time-to-market and engineering costs associated with AI deployment.

The technology achieves this portability through four specialized backends that generate the necessary code for different hardware architectures:

• OpenCL C: For broad compatibility across various devices.
• NVIDIA CUDA PTX assembly: Specifically for NVIDIA GPUs.
• SPIR-V binary: For modern graphics and compute APIs.
• Metal Shading Language: Optimized for Apple Silicon and macOS environments.

Hardware Support and Compatibility

One of the most significant advantages of TornadoVM 4.0 is its extensive hardware reach. It targets a wide array of accelerators, including:

• Dedicated GPUs: NVIDIA, AMD, and Intel.
• Integrated GPUs: ARM Mali, Intel HD Graphics, and Apple Silicon.
• FPGAs: Xilinx and Intel.
• Multi-core CPUs: Standard processor acceleration.

To ensure stability and performance, TornadoVM 4.0 requires JDK 21 (or a GraalVM distribution based on JDK 21). It maintains broad JVM support, working seamlessly with OpenJDK, GraalVM, Eclipse Temurin, Red Hat Mandrel, Amazon Corretto, Microsoft OpenJDK, and Azul Zulu.

Bringing AI to Java: GPULlama3.java

The intersection of Java and AI is most evident in the introduction of GPULlama3.java. This is the first Java-native implementation of Llama3 that automatically compiles and executes Java code on GPUs. By removing the require for complex wrappers or external C++ libraries, TornadoVM allows Java developers to execute AI workloads faster and more cost-effectively.

Installation and Deployment

TornadoVM is distributed via the official website and SDKMAN!. Developers can install a version tailored to their specific OS, architecture, and backend (such as 2.2.0-opencl, 2.2.0-ptx, or 2.2.0-full for all backends). Once installed, the tornado --devices command allows users to verify that the system detects the available CPU/GPU accelerators and that all required drivers are correctly configured.

Frequently Asked Questions

Does TornadoVM require me to write CUDA or OpenCL code?

No. TornadoVM automatically generates the necessary OpenCL, PTX, SPIR-V, or Metal code from your Java programs, allowing you to stay within the Java ecosystem while gaining hardware acceleration.

Which JVMs are supported?

TornadoVM supports a wide range of JVMs, including OpenJDK, GraalVM, Eclipse Temurin, Red Hat Mandrel, Amazon Corretto, Microsoft OpenJDK, and Azul Zulu.

Can TornadoVM run in the cloud?

Yes. TornadoVM supports deployment on Amazon AWS across CPUs, GPUs, and FPGAs, and it can also be deployed using Docker containers for GPU workloads.

Looking Forward

As AI workloads become central to enterprise software, the ability to leverage heterogeneous hardware without leaving the Java environment is a critical evolution. By bridging the gap between high-level Java development and low-level hardware acceleration, TornadoVM 4.0 positions Java as a viable, high-performance language for the next generation of AI-ready software.