TornadoVM 7 lets Java developers call CUDA libraries directly from their code
TornadoVM 7 Enables GPU-Native Workloads from Java Applications
TornadoVM 7.3 is a major milestone for the open-source framework that accelerates Java programs on multi-core CPUs and GPUs without replacing the JVM. The release swaps JNI for the Foreign Function & Memory API, drops the experimental JVMCI dependency, and adds an optimized CUDA backend with cuTile support plus direct calls to cuBLAS and CUTLASS from TaskGraphs. Multi-GPU execution and Apple MXL support also arrive, alongside the jitLLM 2.0 inference engine.
The idea is that you can develop locally on your laptop and use the hardware available locally, and then deploy into a server that has Nvidia GPUs, and achieve major performance improvements - a form of "Write Once, Run Anywhere" for modern ML workloads.
- mikepapadim
https://github.com/beehive-lab/TornadoVM this is main repo.
You can use it with any JDK21+ and Nvidia or Apple hardware and their Nvidia libs or MLX libs.
- the_arun
Does this mean we can use Java to build AI infrastructure?
- amir734jj
Basically JAX for Java instead of Python