Zero Code Change GPU-Powered Graph Analytics with NetworkX and cuGraph
summary
NetworkX now dispatches graph algorithms to GPU backends so existing code runs much faster with zero code change. The demo shows enabling dispatching via an environment variable and reusing the same NetworkX calls with nx-cugraph (cuGraph backend). Betweenness centrality on ~3.7M nodes and ~16M edges drops from hours to seconds; Louvain on 1.6M nodes and 30M edges drops from ~38 minutes to ~3 minutes, with graph construction and Leiden runs similarly accelerated.
AI-written summary · about
chapters (8)
- 0:00GPU-powered graph analytics intro
- 1:07Why NetworkX dominates
- 7:30Dispatching with NX cuGraph
- 11:02Social network example
- 18:47Louvain algorithm insights
- 20:10Leiden via backend-on
- 22:14Key takeaways
- 22:57Q&A
related talks
40mRun Large-Scale Graph Analytics Using The Most Popular Graph Analytics Library Available
36mNetworkX is Fast Now- Zero Code Change Acceleration
31mFast NetworkX and How Accelerated Backends Are Changing Graph Analytics
28mNetworkX is Fast Now: Graph Analytics Unleashed
31mGraph-Based Machine Learning with Python
16mAccelerating Distributed Graph Learning by Using Collaborative In-Network...
02h 47mNetwork Analysis Made Simple
19mNeuGraph: Parallel Deep Neural Network Computation on Large Graphs