RAPIDS: GPU-Accelerated Data Science for PyData Users
summary
Claim: RAPIDS makes end-to-end PyData pipelines run on NVIDIA GPUs with cuDF (GPU-accelerated pandas) and cuML / QML (GPU scikit-learn), showing real examples in Colab on a T4. Demonstrations include the zero-code pandas accelerator with profiling (about 94% pandas test coverage reported) and benchmarks like a group-by that drops from ~5s to ~0.4s on GPU. The notebook builds working pipelines, shows integration with Polars and Dask, and reports typical 10–100x speedups when data movement and ops suit the GPU.
AI-written summary · about
chapters (16)
- 0:00RAPIDS and speakers intro
- 1:12GPU basics and CUDA
- 8:17Nvidia GPUs and CUDA
- 13:00Google Colab GPU setup
- 15:29cuDF and GPU ETL intro
- 23:22cuDF filtering and grouping
- 32:15Zero-code cuDF pandas
- 44:24NY parking violations data
- 50:13cuDF profiling insights
- 59:57cuDF with Polars intro
- 1:07:10Polars GPU benchmarks
- 1:14:00cuML and scikit-learn
- 1:21:56Accelerating scikit-learn with cuML
- 1:27:15RAPIDS resources
- 1:30:26RAPIDS roadmap
- 1:34:42Q&A
related talks
16mGPU-Accelerated Data Science for PyData Users
34mRAPIDS: How to accelerate your data science pipeline by orders of magnitude
38mRAPIDS- Open Source GPU Data Science
01h 22mHigh-Performance Data Science at Scale with RAPIDS, Dask, and GPUs
01h 12mGPU-Accelerated Data Analytics in Python
20mEnd-to-End Data Science on GPUs
39mAccelerating Data Science with RAPIDS
41mAccelerating Data Science with RAPIDS