Speed up Python with hardware accelerators
summary
Benchmarks compare plain Python loops to NumPy, Numba and TensorFlow with concrete timings: 100x100 matrix multiply and pairwise-distance tests show orders-of-magnitude speedups; scipy.spatial.cdist is unexpectedly fast on CPU; a Monte Carlo integrator and a PDF interpolation pipeline (1M points) reach almost 100× versus C++ LHAPDF when using TensorFlow on GPU; examples and code live on the GitLab repo and the rule is vectorize and offload independent high-count work to Numba JIT or GPU.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 1:53Python performance basics
- 10:55Python slowness causes
- 11:17Matrix multiplication example
- 12:42Numba and TensorFlow acceleration
- 15:30Pairwise distance benchmark
- 16:57Monte Carlo integration
- 18:23Complex interpolation case
- 21:36MadFlow tool comparison
- 23:26Quantum computing acceleration
- 24:48Conclusions and recommendations
- 25:31Q&A
related talks
02h 55mGPU Accelerated Python w: CUDA
35mPython at the Speed of Light: A Complete Tour of the CUDA Ecosystem
45mPython for Speed, Scale, and Science
30mHighLoad Python SIMD, GPU, and Horizontal Scaling Looking into Silicon
30mCUDA in Your Python: Effective Parallel Programming on the GPU
44mPython's Bright Future in Science
33mWhen GPUs Make Python Slower: Patterns and Pitfalls
01h 29mGPU development with Python 101