Up and Away: JDK Optimizations
summary
Intel and Oracle report concrete JDK 10/11 wins: vector API (JEP 338) and auto-vectorizer changes enable AVX/AVX‑512 code paths and FMA use, yielding up to ~6x speed for vectorized sqrt, ~7x for floats, ~2x for array/string mismatch, and ~70% gains moving AVX2 intrinsics to AVX‑512. Class-data-sharing cuts startup ~15–20%. New JVM flags let the heap or old‑gen live on persistent memory and direct I/O / RDMA support reduce kernel time. Many changes land in JDK 11; try the flags and Panama branch code.
AI-written summary · about
chapters (16)
- 0:00Introduction
- 1:24OpenJDK Community - working together to improve Java
- 2:05Features refresh: JDK 9
- 5:47Features refresh: JDK 10
- 10:21Features refresh: JDK 11 (cont)
- 11:28JDK 10 and 11 x86 optimizations in details
- 12:40Optimization for small element vectors
- 13:37Array Compare Optimization
- 15:03Improve intrinsics performance by using AVX512
- 15:43FMA (fused multiply-add)
- 17:16FMA Vectorization
- 19:08Whole Java heap on PM
- 22:03Vector API
- 29:08Future Features: JDK 12 (cont)
- 30:53Future Features (cont.)
- 36:24Call For Action
related talks
48mJust-In-Time Compilation for Java Performance: Recent and Ongoing Improvements
51mJava at Speed: Getting the Most Out of Modern Hardware
51mJava Performance Update: From JDK 21 to JDK 25
51mAccelerating performance of Java applications on Arm64
48mJava 9 Performance
44mHotspot JVM Auto Vectorization and Vector API
56mJDK 9 & 10: Mission Accomplished, What Next For Java
53mA year with jdk11 in production.