High-Performance Data Science at Scale with RAPIDS, Dask, and GPUs
summary
A concrete claim: RAPIDS + Dask + GPUs move entire ETL, ML and graph pipelines onto GPU memory to avoid copies and give huge real results. Tools and bridges are explicit: cuDF dataframes, cuML models, UCX for fast communication and DLPack for zero‑copy interop. Benchmarks show CSV parsing that runs ~1.75s vs ~20s in pandas and typical speedups of 10x to >1000x; multi‑GPU random forests and a PageRank run on 16 billion edges PageRank and large graphs finish in seconds, so the components actually work end‑to‑end.
AI-written summary · about
chapters (19)
- 0:00High-performance data science intro
- 1:24CPU limits in-memory
- 3:30Why GPUs excel
- 5:15GPU memory sharing
- 6:18GPU pipelines benefits
- 8:45Python data science overview
- 10:09Scaling with Dask & GPUs
- 14:46cuDF for GPU ETL
- 18:23GPU string processing & I/O
- 22:54NumPy to GPU arrays
- 26:04cuML machine learning
- 30:29cuGraph graph analytics
- 33:44Geospatial acceleration
- 36:18RAPIDS integrations
- 39:12RAPIDS getting started
- 42:22cuDF benchmarks demo
- 57:53Hardware cost efficiency
- 1:01:17Deep learning interoperability
- 1:11:46Dask & cuDF at scale demo
related talks
38mRAPIDS- Open Source GPU Data Science
41mAccelerating Data Science with RAPIDS
34mRAPIDS: How to accelerate your data science pipeline by orders of magnitude
20mEnd-to-End Data Science on GPUs
39mAccelerating Data Science with RAPIDS
01h 08mGPU Data Science with Rapids
01h 36mRAPIDS: GPU-Accelerated Data Science for PyData Users
01h 12mGPU-Accelerated Data Analytics in Python