RAPIDS: How to accelerate your data science pipeline by orders of magnitude
summary
RAPIDS is a suite that runs end-to-end data science on GPUs using cuDF, cuML, cuGraph and Dask with Apache Arrow and UCX plumbing, offering drop-in replacement for pandas and scikit-learn style APIs. Examples show a 2 GB CSV load ≈16× faster, merges and joins often 500–1000× speedups on joins/merges, PCA dropping from 57 s to 4.3 s, xgboost on 16 GPUs beating 100 CPU servers ≈10×, and single-GPU graph analytics handling 500 million edges on one GPU with multi-GPU scaling.
AI-written summary · about
chapters (10)
- 0:00Welcome
- 1:27What is RAPIDS
- 3:35RAPIDS integration
- 7:05RAPIDS technical foundation
- 12:03cuDF introduction
- 16:20cuDF benchmarks
- 22:43cuML introduction
- 26:35cuML benchmarks
- 31:16RAPIDS graph tools
- 31:37Q&A
related talks
38mRAPIDS- Open Source GPU Data Science
41mAccelerating Data Science with RAPIDS
01h 22mHigh-Performance Data Science at Scale with RAPIDS, Dask, and GPUs
39mAccelerating Data Science with RAPIDS
01h 36mRAPIDS: GPU-Accelerated Data Science for PyData Users
01h 08mGPU Data Science with Rapids
01h 12mGPU-Accelerated Data Analytics in Python
20mEnd-to-End Data Science on GPUs