airhacks.fm podcast with adam bien Babylon and java.util.json
Oct 26, 2025
Paul Sandoz, a Java platform engineer at OpenJDK, discusses the exciting evolution of Java, touching on energy efficiency and performance improvements. He dives into Project Babylon and its integration with Vector API, enabling advanced AI capabilities on the JVM. Listeners will learn about TornadoVM's GPU optimizations, Python's enterprise challenges, and the introduction of a minimal JSON API for OpenJDK. Paul shares insights on pattern matching and the benefits of a lightweight API, making Java a more competitive option for enterprise and embedded systems.
AI Snips
Chapters
Transcript
Episode notes
Java's Silent Performance Evolution
- Java's platform improvements have steadily boosted performance, memory use, and code generation over many releases.
- Paul Sandoz and Adam Bien note Java 25's advances make it far more competitive than older studies suggest.
Architecture Drives Compiler Progress
- Hardware-specific compiler work (ARM, SVE2, RISC-V) drives an ongoing optimization arms race across runtimes and compilers.
- OpenJDK receives architecture-specific contributions (NVIDIA, Intel) to exploit new CPU/GPU instructions.
TornadoVM's GPU Inference Breakthrough
- TornadoVM researchers moved data pointers to keep tensors on the GPU and avoid transfer overhead, yielding large inference gains.
- Adam Bien reports 6–10x speedups on NVIDIA and ongoing Metal work for Apple Silicon.