A Beginners Guide to GPUs for Pythonistas
summary
Shows concrete timings and tools: pandas 156 ms vs cuDF 3.56 ms for a mean, and a merge that goes from pandas 70 s vs cuDF 0.367 s, demonstrating large, measured speedups. Explains the GIL constraint and why GPUs with thousands of cores help parallel workloads, describes an open‑source GPU dataframe (cuDF) in RAPIDS, and points to a Google Colab RAPIDS notebook and local GPU runs so the code and notebooks can be reproduced.
AI-written summary · about
chapters (8)
- 0:00Introduction to speaker
- 1:04Gpus and their purpose
- 3:20Cpus vs Gpus explained
- 7:35Addressing Python performance
- 11:54The global interpreter lock
- 16:00Pandas versus Python
- 20:59Accelerating data with rapids
- 26:28Accessibility and conclusion
related talks
02h 55mGPU Accelerated Python w: CUDA
01h 12mGPU-Accelerated Data Analytics in Python
02h 28mGPU Accelerated Python
27mThe art of wrangling your GPU Python environments
31mcuDF: RAPIDS GPU-Accelerated Dataframe Library
01h 36mRAPIDS: GPU-Accelerated Data Science for PyData Users
33mWhen GPUs Make Python Slower: Patterns and Pitfalls
29mcudf.pandas: The Zero Code Change GPU Accelerator for Pandas