Building a Data-Oriented Future
summary
Unity ships a concrete stack: a job scheduler plus Burst AOT compiler compiling an HPC# subset, native memory containers (no GC) and an ECS with 16 KB chunks and archetypes, all designed to run wide across cores. Claims: data layout and aliasing cut typical cache waste (the talk quantifies ~90% wasted cache lines) and deliver editor iteration under 500 ms. Result: multi‑threaded code becomes verifiable and incremental instead of an expert‑only rewrite, enabling optimization paths from prototype to production.
AI-written summary · about
chapters (24)
- 0:00Introduction
- 1:00Why Unity
- 2:30My Mission
- 3:30What I do at Unity
- 4:20The default approach
- 5:10Three big lies
- 7:05No preproduction wall
- 9:30Core tech
- 11:25Verification of correctness
- 12:00Feedback on correctness
- 13:05Why HPCSharp
- 13:45Why CSharp
- 14:45Why Burst
- 16:00Editor Inspector
- 17:55Structure Conversions
- 18:25Code Unity
- 19:20What is ECS
- 21:10What is Chunk
- 21:45What is Work Kernel
- 23:20Global Energy
- 24:55Data Transformation
- 26:40Understanding the Data
- 28:10Generic Frameworks Platform Independent Future Proofing
- 31:30Hardware
related talks
17mKeynote Analysis
16mEmpowering Fearless Data Engineering
31mWeld: An Optimizing Runtime for High Performance Data
36mRay Data: What's New - GPU Shuffle, Unity Catalog, DataSourceV2
58mComposable Data Systems, the new frontier
32mStreamlining Data Access for with
20mKeynote – Ion Stoica
21mMoore’s Law And The Future Of Big Data