[nx-cugraph] Calling All NetworkX Users Looking for GPU Acceleration

[Ralph Liu US]

Hello all,

In conjunction with the great members of the NX community, we developed `nx-cugraph`, a NetworkX backend that uses RAPIDS cuGraph and NVIDIA GPUs to significantly improve algorithm performance. This GPU-powered backend allows users to gain significant speed-ups while maintaining the same API.

We're looking for projects that could benefit from GPU acceleration. If your NetworkX-based workflows are running into performance bottlenecks, please let us know—we'd love to work with you to help speed things up where possible.

Are you already using `nx-cugraph` to speed up NetworkX but want support for specific features? e.g., adding a missing algorithm, fixing a conversion bottleneck, etc. Let us know what else you'd like to see added in the future.

We'd love to connect with more users and get feedback. Feel free to reply here or reach out directly. Looking forward to collaborating with you all.

Kindly,
Ralph Liu
ral...@nvidia.com

Links to the project:

Landing Page: https://rapids.ai/nx-cugraph/

Docs: https://docs.rapids.ai/api/cugraph/stable/nx_cugraph/
GitHub: https://github.com/rapidsai/nx-cugraph/

[Russell Jurney]

I think this is really cool and I'm sure to use it. Regarding GPU accelerated graph processing... I'm interested in doing GPU accelerated network motif matching. At the moment my plan is to use Spark-RAPIDS and GraphFrames motif finding.

Can you accelerate motif mining with networkx / nx-cugraph backend?

Thanks,

Russell

--
You received this message because you are subscribed to the Google Groups "networkx-discuss" group.
To unsubscribe from this group and stop receiving emails from it, send an email to networkx-discu...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/networkx-discuss/20eaa76b-d963-455e-8480-4f7d93c9a2b6n%40googlegroups.com.

[Ralph Liu US]

Hi Russell,

By Motif Mining, are you referring to searching a large graph for small recurring sub-graph patterns (e.g., triangles, stars, cliques, etc)?

If so, the nx-cugraph backend is suitable for this application. May I ask what algorithms you're interested in using? So far, I'm aware of louvain_communities and triangles both being supported by the GPU backend. Additionally, we're considering adding support for sub-graph isomorphism algorithms (VF2, VF3) after receiving some community interest.

Looking forward to continuing this discussion.

Kindly,

Ralph

[Russell Jurney]

I'm referring to graphlet or subgraph matching in a larger graph - yes, an isomorphism test - ideally as compared with a random network null model - but in this context probably just matching. It would be awesome if there as GPU accelerated subgraph matching in networkx. VF2 an VF3 support would be amazing! If there were subgraph matching that included node and edge labels - heterogeneous networks - that would be even better. Including properties - better than that :) One approach to that is neural subgraph matching - and that would take the taco! It might be a way to include properties in an isomorphism test... to some extent you can transform a labeled property graph into a heterogeneous network.

Russell

To view this discussion visit https://groups.google.com/d/msgid/networkx-discuss/0a321831-1a4c-4b4c-a60b-a529801cdc8fn%40googlegroups.com.