NetworkX is Fast Now: Graph Analytics Unleashed
summary
NetworkX adds a dispatch layer (since NetworkX 3.0) that runs algorithms on heterogeneous backends and shows concrete speedups: a 90k‑edge / 3k‑node clustering run drops from ~25 seconds to ~0.5 seconds on nx‑cugraph, benchmarks on a Quadro RTX 8000 report up to 10_000+× speedup and cached conversions give 700×–750× for repeated runs. Implementations include nx-cugraph / GraphBLAS backends, conversion caching stores GPU graph objects in‑process, and some algorithms still cost conversion overhead.
AI-written summary · about
chapters (15)
- 0:00Welcome
- 0:23Speed transformation example
- 1:51Performance in graph analytics
- 2:55NetworkX in open-source tools
- 3:37Graphs for data insights
- 5:50NetworkX as analysis library
- 8:23Scaling Python algorithms
- 9:50Dispatching backend system
- 13:04GPU acceleration with nx-cugraph
- 17:57Benchmarking speed improvements
- 18:41Optimizing storage formats
- 19:24NetworkX as graph database
- 21:11Future plans and community
- 22:54Backend scalability with nx-cugraph
- 24:01Q&A on backend compatibility
related talks
40mRun Large-Scale Graph Analytics Using The Most Popular Graph Analytics Library Available
27m100x Faster NetworkX: Dispatching to GraphBLAS
32mZero Code Change GPU-Powered Graph Analytics with NetworkX and cuGraph
29mUnderstanding API Dispatching in NetworkX
24mNetworkX - your Unexpected Assistant for Clustering Analysis
02h 47mNetwork Analysis Made Simple
16mAccelerating Distributed Graph Learning by Using Collaborative In-Network...
39mAccelerated Data Science Libraries