2025-11-02 –, Hall 1
CPython drops the GIL! Dive into NoGIL, green threads, and async — with real benchmarks vs. multiprocessing & other languages. Learn when free-threading gives a boost, where it doesn't, and how to migrate your code step-by-step.
2025 has been a turning point for Python in high-load systems. CPython 3.14 removes the GIL and introduces free threading (FT)—the biggest performance leap in 20 years. But is Python truly ready to compete with Go and Java on CPU-intensive workloads?
Using live benchmarks, I’ll show where FT outpaces multiprocessing by 2–3× (ML pipelines, large DataFrames) and where it falls short. We’ll dissect NUMA effects across 500 RL environments and look at GPU orchestration. We’ll compare against Java/Go on identical tasks. I’ll demo adaptive specialization in action—how CPython learns from your code at runtime. We’ll discuss the JIT and the new tail-calling interpreter. We’ll analyze flame-graph profiles and pinpoint bottlenecks.
I’ll show how much faster CPython has become: compare 3.9/3.12/3.14 (3.15) and make the case that production should migrate to 3.12, which is in official bug-fix support.
Practical takeaways:
• A migration checklist for FT with a risk map of potential performance regressions
• Source code for building CPython without the GIL + build instructions
• A benchmark suite for testing your high-load cases
• A profiling toolkit: ready-to-use perf record + flame-graph scripts
• Clear architectural decision criteria: FT / multiprocessing / asyncio / green threads / other languages
• Anti-patterns: when FT is definitely the wrong choice
Specializes in CPython internals, optimization, and high-performance computing. Driven by GPU acceleration and CPU vectorization. Evolved from ML systems to CPython core research engineer. 8+ years leading teams in AI, maths, and physics. PyCon speaker. Lecturer at Moscow Institute of Physics and Technology – top 1 Russian university. Open to talks and collaboration.