Run Large-Scale Graph Analytics Using The Most Popular Graph Analytics Library Available
summary
NetworkX adds backend support and an NVIDIA backend nx-cugraph implements cuGraph so existing NetworkX code runs on GPU by setting an environment variable. Demonstrations show minute‑scale runs dropping to seconds (example: ~8 minutes to ~5 seconds; PageRank from hours to <2 seconds), about 60 algorithms available, and large graphs (order 100k nodes to millions of edges) benefiting most. Some algorithms lag due to Python↔GPU conversion overhead, multi‑GPU via Dask and further optimizations are planned under the zero-code change acceleration approach.
AI-written summary · about
chapters (14)
- 0:00GPU accelerated NetworkX intro
- 1:27What is GPU accelerated NetworkX
- 3:34RAPIDS and GPU acceleration
- 6:04Zero code change with NetworkX backend
- 7:31Who benefits from GPU acceleration
- 8:35Performance benchmarks
- 10:42Workflow changes with GPU
- 13:58Recommendation systems case study
- 17:05Performance comparison to NetworkX
- 19:40How GPU acceleration works
- 23:16Zero code vs explicit dispatch
- 26:29Debugging and optimization
- 27:10Future roadmap
- 29:16Q&A
related talks
31mFast NetworkX and How Accelerated Backends Are Changing Graph Analytics
28mNetworkX is Fast Now: Graph Analytics Unleashed
32mZero Code Change GPU-Powered Graph Analytics with NetworkX and cuGraph
19mNeuGraph: Parallel Deep Neural Network Computation on Large Graphs
31mGraph-Based Machine Learning with Python
37mGraphs, Networks and Python: The Power of Interconnection
02h 47mNetwork Analysis Made Simple
56mGraph Analytic Systems