End-to-End Data Science on GPUs
summary
RAPIDS implements an end-to-end GPU data science stack with cuDF for GPU DataFrame ETL, DLPack/CUDA array interface for zero-copy handoff to PyTorch/TensorFlow, and Dask + UCX over InfiniBand for scale-out. The team ports TPCx-BB end-to-end to GPUs and reports a ~10x runtime win over the best public CPU solution and ~5x lower total cost, single-node cases showing ~40x cost or ~51x speed improvement. Microbenchmarks claim up to ~970x for joins, ~300x for groupbys, ~700x SVM, and ~23–36x forest inference.
AI-written summary · about
chapters (7)
- 0:00Introduction
- 1:03Why accelerate data science with GPUs
- 4:17RAPIDS
- 10:00KUMAL
- 14:45Distributed Compute
- 17:15Benchmarks
- 18:49Resources
related talks
01h 22mHigh-Performance Data Science at Scale with RAPIDS, Dask, and GPUs
38mRAPIDS- Open Source GPU Data Science
41mAccelerating Data Science with RAPIDS
34mRAPIDS: How to accelerate your data science pipeline by orders of magnitude
01h 08mGPU Data Science with Rapids
01h 36mRAPIDS: GPU-Accelerated Data Science for PyData Users
39mAccelerating Data Science with RAPIDS
01h 12mGPU-Accelerated Data Analytics in Python