From CPU to GPU and FPGAs: Supercharging Java Applications with TornadoVM
summary
TornadoVM compiles annotated Java methods to OpenCL, PTX or SPV via Graal and dispatches them with oneAPI Level Zero, yielding real results: a 500x microkernel gain when calling vendor libraries, up to 240x speedup vs Java seq, and an example image pipeline dropping from ~8s to 0.3s on RTX 2060 after JIT. Runtime features include a task graph + execution plan, automatic batching, multi-device concurrent execution and live task migration; it integrates with Flink and plans Panama support.
AI-written summary · about
chapters (5)
- 0:00Start
- 1:07Outline
- 24:21Batch Processing for Big Data Workloads
- 25:11Multi-Device Execution for Single TaskGraphs
- 26:22Integration with Big Data Platforms Flink [Experimental]
related talks
46mTornadoVM: Supercharge your Java with GPU acceleration
51mTornadoVM: Transparent Hardware Acceleration for Java…and Beyond!
44mWrite Once, Accelerate Everywhere: GPU-Ready Java with TornadoVM
41mTornadoVM
43mHow to make your electrons do more in Java?
49mJVM in the Age of AI: Babylon, Valhalla, TornadoVM and friends
40mHow to make your electrons do more in Java
33mOptimizing Java on the EU processor platform