TornadoVM

TornadoVM 8.0.0 is out! After 10 years: thank you, Graal. Java kernels are now compiled by TC, our own pure-Java compiler to CUDA, OpenCL and Metal. • AOT ready • No JVMCI • Now Apache 2.0 • Your code doesn't change 👉 www.tornadovm.org/blogs/tornad... #OpenSource #Java #AI #GPU

Great to see @tornadovm.org and @jitllm.io hit #1 on Hacker News!” Check out the latest article on @infoq.com by @ogkittylyst.bsky.social : www.infoq.com/news/2026/10...

Had so much fun at #JDD 2026 in Kraków yesterday. I presented “Java Goes CUDA-Native”: new @tornadovm.org features that let @jitllm.io run LLM inference at GPU-native speed, from Java. tornadovm.org jitllm.io #opensource #Java #TornadoVM #GPU #AI

Yesterday my colleagues @maryxek.bsky.social and @mikepapadim.bsky.social showed live demos about using NVIDIA cuFFT and cuBLAS with @tornadovm.org for signal filtering and @jitllm.io! NVIDIA performance is a jar away from Java. #Java #CUDA #TornadoVM

With the latest #TornadoVM version you can accelerate end-to-end Apache #Flink pipelines on the GPUs with native #LLM inference. No Python, no external APIs, all in one deployable jar!

Optimized GPU libraries (e.g. cuBLAS, cuDNN, etc.) by NVIDIA, now one jar away for your Java - without caring about JNI. We also show demos with Apache Flink and @jitllm.io. Watch our talk in @devoxx.com 👇 youtu.be/6SPnBmQrYEk?... #opensource #AI #TornadoVM #Java #GPU

jitLLM is part of the @langchain4j.dev 1.21.0 release! 🚀 Read more here: docs.langchain4j.dev/integrations... #opensource #LLM #GPU #Java #LangChain4j #Inference

Our latest addition to the TornadoVM family. Blazing fast AI inference at native speed!

Can Java compete with llama.cpp/ollama? YES! With @jitllm.io! jitLLM uses NVIDIA CUDA features and libraries (cuBLAS, cuDNN) and reaches 0.90× to 1.12× of llama.cpp's throughput! jitLLM is empowered by @tornadovm.org and will be soon in @langchain4j.dev and @quarkus.io. github.com/beehive-lab/...

Next week we will be in @devoxx.com and speak about @tornadovm.org, @jitllm.io and how we accelerate Java on GPUs, using #NVIDIA native libraries! Come and talk to us, to learn more! #opensource #Java #AI #TornadoVM #jitLLM #GPU

TornadoVM 7.0.0 is out! New TileContext API: Write #GPU kernels from #Java in tiles, not threads, via #NVIDIA #cuTile model. Now, cuTile methods can be mixed with cuBLAS functions in one TaskGraph, optimizing data transfers. 👉 github.com/beehive-lab/TornadoVM/releases/tag/v7.0.0 #opensource #AI

Java 27 arrived yesterday! And we are compatible since Day-0! #opensource #Java #TornadoVM #GPU

New #TornadoVM release, and it's JVMCI-free. 📌No more separate jdk25 branch, no back-syncing every PR from develop - one build for JDK 21–27. 📌Native code also migrated from JNI to Panama's FFM API - 7.5k lines of C++ gone. Kudos to all contributors! 👏 #opensource #Java #GPU #AI

TornadoVM is now modernized and forwards compatible with any JDK > 21 while receiving significant performance boosts! Can't wait to see those benefits translating to our AI inference engine in our next release!

TornadoVM 6.0.0 is here - JVMCI-free and JNI-free. ✅ Compatible with any JDK 21-27 distribution. ✅ Cold start down 37%. ✅ 4.1× faster GEMM with Tensor-Core MMA. ✅ ~7,400 lines of C/C++ glue deleted. 📖 New blog: www.tornadovm.org/blogs/tornad... #Java #GPU #CUDA #TornadoVM #OpenSource #HPC #AI

The next TornadoVM release 6.0 will remove completely the dependency to JVMCI and will be onwards compatible. Babylon/HAT are complementary and we may integrate some parts when they are stable. Also we will soon land a PR to integrate with SpringAI! Your feedback is greatly appreciated!

Great news! Three talks accepted at @devoxx.com 2026! Hope to see you all and discuss the future of #Java and #AI!

3 sessions accepted at @devoxx.com Belgium! 🛠️ Tools in Action: NVIDIA ecosystem from Java via #TornadoVM's Hybrid API 🎤 Talk: Building LLM inference engines in Java with TornadoVM 🧑‍💻 Hands-on Lab: From install to insight — a GPU lab for Java devs Stay tuned #Java #CUDA #LLM #AI #GPU #opensource

Our #Java #AI inference engine just got another major update with a huge performance leap and native #CUDA integration!

Great effort and what an achievement! #LLM inference implemented in #Java, automatically accelerated on #GPUs with #TornadoVM & CUDA native support! #opensource #AI #Java #GPU

GPULlama3 v1.0.0 is out - now #CUDA-native ⚡ ✅ TornadoVM CUDA backend w/ tensor-core (MMA) batch prefill ✅ FP16 & Q8_0 · Llama, Qwen, Devstral & more models ✅ An OpenAI-compatible server (llama-tornado --server) Pure #Java. No JNI. github.com/beehive-lab/GPULlama3.java #opensource #AI #LLM #GPU

The final release of the AERO software stack is here! Our new v2.0 leaflet showcases how we've optimized cloud deployments on top of the EU processor designs & accelerators. 📄 Get the brochure: aero-project.eu/posts/aero-v... #OpenSource #cloud @hipeac.bsky.social @euprojectpulse.bsky.social

JVM Weekly by @askowronski.bsky.social features @tornadovm.org 5.0.0 release that brings native CUDA support (cuBLAS, cuDNN, etc.) via the new Hybrid API. Same hardware-agnostic Java, can be extended for AMD & Apple Metal libraries. www.linkedin.com/pulse/java-g... #Java #GPU #TornadoVM #opensource

It’s been a busy week in the #Java ecosystem ➤ The GA release of TornadoVM 5.0; point releases of JHipster, Keycloak and Google ADK; maintenance releases of GraalVM Native Build Tools and Micronaut; the OmniFish Build of Payara and introducing Vidocq, and more. 🔗 #InfoQ News Roundup: bit.ly/4pq19Bl

TornadoVM has officially transitioned into a new phase. One line of code, > 150x performance. Becoming #CUDA native unlocks new potentials for high performance #Java #AI inference with zero dependencies. Checkout our latest release, blogs, and new website 👇

TornadoVM 5.0.0 is CUDA-native. One line of #Java. 74× faster. TornadoVM's Hybrid API drops a native cuBLAS call straight into a TaskGraph - same buffers, same #CUDA stream, zero host round-trips. ⚡ Run with the Hybrid API: bit.ly/4vqL7sq 📖 Full engineering deep dive: bit.ly/44VecBf #AI #GPU

The talk is now online! GPU-accelerated, local LLM inference in pure Java, from a #Quarkus #LangChain4j app. No Python. No CUDA. #TornadoVM 🔗 youtu.be/9YCRUwr7pO8 #horizoneurope @hipeac.bsky.social @eucloudedgeiot.bsky.social @quarkus.io @langchain4j.dev

New talk recording from @jprimeconf.bsky.social! Run GPU-accelerated, local LLM inference in Java - from a @quarkus.io @langchain4j.dev app. No Python. No CUDA. Just Java. All inside the JVM - powered by #TornadoVM + GPULlama3. 🔗 youtu.be/9YCRUwr7pO8?... #Java #GPU #LLM #AI #opensource

Final month of #AERO - and we're not slowing down! Run GPU-accelerated LLMs in pure Java from a @quarkus.io #Langchain4j apps. No Python. No CUDA. Just #Java with @tornadovm.org. A collab between @manchester.ac.uk & Red Hat! 🔗 aero-project.eu/posts/runnin... #Java #LLM #AI #LLM #GPU #opensource

Last week I was in Sofia for @jprimeconf.bsky.social! I presented our latest work on @tornadovm.org and #GPULlama3 for seamless #GPU acceleration of local #AI inference with @quarkus.io and @langchain4j.dev. Thanks to organizers, speakers, attendees & photographers for a great event. #opensource

Excited to attend #jPrime2026 in Sofia this week! Looking forward to sharing our work on #GPULlama3, @tornadovm.org and their integration with @quarkus.io @langchain4j.dev. Join me on Thursday (4/6) at 14:05, in Hall A: jprime.io/agenda/275 #jPrime26 #Java #GPUs #AI #TornadoVM

This week, Mary Xekalaki @maryxek.bsky.social presents the latest work at #ICPE2026! “Pulse: A Profiling & Visualization Infrastructure for Heterogeneous Managed Systems” 📅 Thu, May 7 🕒 14:40–15:00 🔗 github.com/beehive-lab/... 📄 zenodo.org/records/1895... #opensource #Java #GPU #TornadoVM

TornadoVM team at @devoxxgreece.bsky.social Great discussions, insightful sessions, and a fantastic community pushing the boundaries of high-performance #Java and #GPUs. 📸 Photo credit: @ddoutsiopoulos.bsky.social

Great to see this work from the TornadoVM team! As part of the AERO project, we’re excited to support open-source innovation that pushes the boundaries of running local LLMs on GPUs directly from #Java. @euprojectpulse.bsky.social @hipeac.bsky.social @eucloudedgeiot.bsky.social #opensource

Last week we gave two presentations at #OCX26 and @devoxxgreece.bsky.social - and the recordings are now available, scan the QR code! From open-source evolution to running LLMs on GPUs directly from Java - there’s a lot to explore. Stay tuned for more! #opensource #Java #TornadoVM #AI #LLM #GPUs

#LangChain4j 1.14.0 is out! 🎉 github.com/langchain4j/...

TornadoVM is featured in the latest Java Pro Magazine special issue (“Java in the Age of AI”, April 2026). 📌https://javapro.io/wp-content/uploads/2026/04/JAVAPRO_04-2026.pdf#page=42 @hipeac.bsky.social @euprojectpulse.bsky.social @eucloudedgeiot.bsky.social #horizoneurope #opensource #community

TornadoVM is featured in the April 2026 issue of Java Pro Magazine - “Java in the Age of AI”! 👉 javapro.io/wp-content/u... In this article, @mikepapadim.bsky.social shared our experiences on building #GPU accelerated #LLM inference libraries with #TornadoVM. #Java #community #AI

We are at #OCX26 this week! Yesterday, Thanos Stratikopoulos from University of Manchester presented the #opensource practices used in @tornadovm.org project for resulting into a production ready software! Stay tuned for more! @euprojectpulse.bsky.social #opensource #cloud #GPU #Java #TornadoVM

On my way to Brussels 🇧🇪 for #OCX2026 “TornadoVM: From Academia to Open Source Success” 
🕓 Tomorrow, 16:00–16:45 | 📍 Studio 4 👉 www.ocxconf.org/event/2026/a... Let’s connect if you’re there! @tornadovm.org #OpenSource #Java #TornadoVM

Great to see external contributions driving real progress! Native #GPU support on Apple Silicon with Metal unlocked by #TornadoVM has enabled running Devstral 2 models directly on-device — a big step forward for performance and accessibility. This is exactly what makes #opensource so powerful!

I have just migrated a Python launch/configuration script to Java 25, and I have asked an LLM to review it 👇

🚀 Gemma4 Java runner: • Single file, no deps • E2B→31B + MoE • GGUF + quant (F16→Q8) • Vector API ⚡ • CLI + thinking modes • GraalVM native + instant TTFT Pure Java 😏 by @mukel.bsky.social github.com/mukel/gemma4...

Our @cern.voxxeddays.ch 2026 talk is now live! We introduce TornadoVM, and show how it’s already used in the ESA #GAIA mission. Highlight: #GPULlama3, a #GPU accelerated #LLM inference in #Java, integrated with @langchain4j.dev + #Quarkus 👉 www.tornadovm.org/gpullama3 📽️ youtu.be/oRVtTEbtt_Y?...

TornadoVM v4.0.0 is out! ✔ Apple Metal backend → native Apple Silicon ✔ SIMD intrinsics for PTX → more performance ✔ CUDA Graphs → reduced overhead ✅ Downloads at TornadoVM → www.tornadovm.org/downloads Big thanks to our contributors! 👏 #Java #JDK25 #JDK21 #TornadoVM #opensource #AI #GPUs

ICYMI our @cern.voxxeddays.ch 2026 talk with @maryxek.bsky.social on @tornadovm.org #Java #GPUs #AI and all👇 youtu.be/oRVtTEbtt_Y?...

We’re excited to see TornadoVM featured at OCX 2026. Thanos Stratikopoulos will share how TornadoVM evolved from academic research into an open source success story. 🕓 16:00-16:45 📍 Studio 4 www.ocxconf.org/event/2026/a... #OCX2026 #TornadoVM #opensource #Java #HPC #research

New paper accepted at #ICPE2026! We present TornadoVMPulse, a profiling & visualization tool for #TornadoVM to analyze and optimize #Java apps on #GPUs. Results for #GPULlama3: ⚡ 18.4% faster execution 📉 38% lower data-transfer cost 🇪🇺 @aero-project.eu 📍 Florence, May26 zenodo.org/records/1895...

The latest version of #LangChain4J supports our latest #GPULlama3 version 0.4.0 for native GPU acceleration of #Java 25 AI Inference!

Join Mary and Michalis at #NVIDIA #GTC to discover how to achieve GPU accelerated Inference for #Java enterprise!