Python Performance Matters
summary
Explains a concrete claim: Scalene is a CPU GPU and memory profiler that reports copy volume (MB/s), separates Python time from native and system time using interrupt-timing, and detects leaks with a sampled donut model. Shows numbers: past profilers slow programs by 2–300x, Scalene runs with ~1.2x–1.3x overhead for memory profiling and ~20–30% on average, spots a 250 MB/s copy hotspot and cuts a run from 30s to 23s. Defaults to ~1.5 MB sampling and handles threads and multiprocessing.
AI-written summary · about
chapters (18)
- 0:00Start
- 1:38Metal Languages
- 7:03Moore's Law
- 7:27Dinard Scaling
- 8:43More Useful Profilers for Python
- 10:59Memory Profiler
- 11:55Profilers
- 14:42The Global Interpreter Lock
- 15:20Multi-Processing
- 16:30Python Is So Slow
- 18:39Copy Volume
- 19:43Gpu Profiling
- 23:07Execution Time
- 24:19Sampling Strategy
- 28:20Automatic Memory Leak Detection
- 30:09Random Sampling
- 31:42Why Is this Useful
- 32:16New Profiler for Python
related talks
26mScalene: A high-performance, high-precision CPU+GPU+memory profiler for Python
15mTriangulating Python Performance Issues with SCALENE
32mProfiling at the speed of light
45mPython for Speed, Scale, and Science
37mTuning Python applications can dramatically increase performance
30mWrite faster Python! Common performance anti patterns
01h 31mDebugging and profiling techniques
30mHighLoad Python SIMD, GPU, and Horizontal Scaling Looking into Silicon