TornadoVM 1.0.4 is out today as the newest version of this solution for [1]Java offloading to GPUs, FPGAs, and other accelerators . TornadoVM allows for nice [2]Java heterogeneous hardware support and with the TornadoVM 1.0.4 brings yet more features.

TornadoVM for device offloading allows targeting OpenCL, NVIDIA PTX, SPIR-V, and more. This OpenJDK and GraalVM plug-in has been designed with a focus on machine learning / deep learning, computer vision, financial software, and other fields. With TornadoVM 1.0.4 there are now Tensor types in the TornadoVM API and interoperability with the ONNX Runtime. TornadoVM 1.0.4 also has added an array concatenation option for native arrays, support for HalfFloat (Float16) in vector types, auto-closable execution plans for automatic memory management, and a number of bug fixes.

TornadoVM 1.0.4 has dropped support for OpenJDK 17 while adding support for OpenJDK 22 as well as GraalVM 22.0.1. SapMachine OpenJDK 21 is also now supported.

Downloads and more details on TornadoVM 1.0.4 via [3]GitHub .



[1] https://www.phoronix.com/news/TornadoVM-0.13

[2] https://www.phoronix.com/news/TornadoVM-1.0

[3] https://github.com/beehive-lab/TornadoVM/releases/tag/v1.0.4