Monitoring Performance on Massive Games.
summary
Live game telemetry moves from average FPS to ms per frame and per-frame breakdowns; each client writes an RPD-style subframe log that compresses from ~1–2 MB to a few hundred bytes for upload. Analytics pipeline (event collectors → Databricks → Tableau, quick JS/Plotly inspection) uses CPU/GPU bucketing (example 8k–9k PassMark bucket) to cut noise, finds real shifts (a 2 ms stall fix shows ~1.9 ms improvement), and exposes hotspots like particle simulation parallelism and region-specific hardware variance. Sample sizes of ~500+ reduce noise.
AI-written summary · about
chapters (15)
- 0:00Start
- 2:26(New) Game Dev Lifecycle
- 10:50A closer look at patch transitions..
- 14:42Filtering by CPU looks like a good option
- 19:03What about per frame perf data?
- 21:12Riot Profile Data File
- 22:53Performance Trends From RPD Data
- 23:30Collect the RPD data
- 23:51Plots for the RPD data
- 24:53Post Process RPD Data
- 25:26Histogram of Game Length
- 26:52Game length vs Frame Time
- 31:28Make it EASY
- 31:57Live games change
- 32:18Ensure Measured Metrics are
related talks
52mOptimising your game AFTER it ships
25mSuperstar Session: How to Spot Performance Problems that Gamers Hate & Even Major Studios Often Miss
42mMaximizing Your Game's Performance in Unreal Engine
58mUE4 Performance and Profiling
47mBenchmarking C++, From video games to algorithmic trading
59mRendering Performance
58mPanel 3: Performance
44mPerformance Analytics Path of Samurai