Accelerating Python With C++
summary
Argues C++ should be paired with Python for speed and shows concrete trades: using pybind11 naïvely makes tiny calls slower (5.7s vs C++ 4.25ms) unless batching or numpy array_t is used (example down to ~62ms); pure-Python sum 64s vs numpy 3s; PyTorch CPU 40s vs GPU 4ms but PCIe round trips cost ~31s for 100 runs; package resolution: pip 4.63s vs uv 0.06s; interactive C++ via CPPY hits ~724ms close to C++ while preserving rapid development.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 2:56C++ and Python usage
- 4:24Python popularity among C++ devs
- 5:51C++ syntax like Python
- 15:09Python performance limits
- 26:11Python performance tools
- 36:15C++ and Python bindings
- 53:18Advanced binding with Syon
- 1:01:16Python from C++
- 1:08:05Dynamic binding with CPPY
- 1:20:59Q&A
related talks
01h 02mWriting Python Bindings for C++ Libraries: Easy-to-use Performance
25m"Boosting Python Numeric Computations with Dynamic C++ Integration"
33mWhy is Python slow?
26mWhy are Python and Ruby slow?
33mA low latency deepdive of Python with Cython
22mExperiences making CPU-bound tasks run much faster
22mPyPy and the future of the Python ecosystem
02h 28mGPU Accelerated Python