HighLoad Python SIMD, GPU, and Horizontal Scaling Looking into Silicon
Talkat PythonAsia 2026 · on Saturday, 2026-03-21· at 14:00· Teresa Yuchengco Auditorium (Main Hall)·
summary
Covers from wafer manufacturing to running Python on silicon and shows practical wins: hyperthreading yields about 10–40% extra throughput by overlapping cache misses, SIMD/AVX with XMM 128‑bit registers lets int8 batches compute roughly eight times more elements per instruction than float64, Numba decorators accelerate a vectorized loop anywhere from ~30× up to 3500× in microbenchmarks, and GPU runs with CuPy or CUDA reach about 60× speed on Colab; demo code and benchmarks compile and run, showing where each tool actually helps.
AI-written summary · about
chapters (10)
- 0:00Introduction
- 1:13Silicon chips production
- 6:14CPU transistors and gates
- 9:52CPU vs GPU architectures
- 12:25Python to hardware
- 14:57Multithreading and cache misses
- 18:33SIMD instructions
- 23:22Accelerating Python with Numba
- 26:36GPU computing with CuPy
- 28:42Closing and Q&A
related talks
02h 28mGPU Accelerated Python
30mCUDA in Your Python: Effective Parallel Programming on the GPU
34mAccelerating GPU Algorithms in pure Python
33mWhen GPUs Make Python Slower: Patterns and Pitfalls
27mSpeed up Python with hardware accelerators
33mLet's Run Python on a Supercomputer!
45mPython for Speed, Scale, and Science
01h 02mExploit your GPU power with PyCUDA (and friends)