How it Works

NetworkX has the ability to dispatch function calls to separately-installed third-party backends.

NetworkX backends let users experience improved performance and/or additional functionality without changing their NetworkX Python code. Examples include backends that provide algorithm acceleration using GPUs, parallel processing, graph database integration, and more.

While NetworkX is a pure-Python implementation, backends may be written to use other libraries and even specialized hardware. nx-cugraph is a NetworkX backend that uses RAPIDS cuGraph and NVIDIA GPUs to significantly improve NetworkX performance.

Enabling nx-cugraph

It is recommended to use networkx>=3.4 for optimal zero code change performance, but nx-cugraph will also work with networkx 3.2+.

NetworkX will use nx-cugraph as the backend if any of the following are used:

NX_CUGRAPH_AUTOCONFIG environment variable.

The NX_CUGRAPH_AUTOCONFIG environment variable can be used to configure NetworkX for full zero code change acceleration using nx-cugraph. If a NetworkX function is called that nx-cugraph supports, NetworkX will redirect the function call to nx-cugraph automatically, or fall back to either another backend if enabled or the default NetworkX implementation. See the NetworkX documentation on backends for configuring NetworkX manually.

bash> NX_CUGRAPH_AUTOCONFIG=True python my_networkx_script.py

backend= keyword argument

To explicitly specify a particular backend for an API, use the backend= keyword argument. This argument takes precedence over the NX_CUGRAPH_AUTOCONFIG environment variable. This requires anyone running code that uses the backend= keyword argument to have the specified backend installed.

Example:

nx.betweenness_centrality(cit_patents_graph, k=k, backend="cugraph")

Type-based dispatching

NetworkX also supports automatically dispatching to backends associated with specific graph types. Like the backend= keyword argument example above, this requires the user to write code for a specific backend, and therefore requires the backend to be installed, but has the advantage of ensuring a particular behavior without the potential for runtime conversions.

To use type-based dispatching with nx-cugraph, the user must import the backend directly in their code to access the utilities provided to create a Graph instance specifically for the nx-cugraph backend.

Example:

import networkx as nx
import nx_cugraph as nxcg

G = nx.Graph()

# populate the graph
#  ...

nxcg_G = nxcg.from_networkx(G)             # conversion happens once here
nx.betweenness_centrality(nxcg_G, k=1000)  # nxcg Graph type causes cugraph backend
                                           # to be used, no conversion necessary

Command Line Example

---

Create bc_demo.ipy and paste the code below.

import pandas as pd
import networkx as nx

url = "https://data.rapids.ai/cugraph/datasets/cit-Patents.csv"
df = pd.read_csv(url, sep=" ", names=["src", "dst"], dtype="int32")
G = nx.from_pandas_edgelist(df, source="src", target="dst")

%time result = nx.betweenness_centrality(G, k=10)

Run the command:

user@machine:/# ipython bc_demo.ipy

CPU times: user 7min 36s, sys: 5.22 s, total: 7min 41s
Wall time: 7min 41s

You will observe a run time of approximately 7 minutes…more or less depending on your CPU.

Run the command again, this time specifying cugraph as the NetworkX backend.

user@machine:/# NX_CUGRAPH_AUTOCONFIG=True ipython bc_demo.ipy

CPU times: user 4.14 s, sys: 1.13 s, total: 5.27 s
Wall time: 5.32 s

This run will be much faster, typically around 5 seconds depending on your GPU.

Note, the examples above were run using the following specs:

    NetworkX 3.4
    nx-cugraph 24.10
    CPU: Intel(R) Xeon(R) Gold 6128 CPU @ 3.40GHz 45GB RAM
    GPU: NVIDIA Quadro RTX 8000 80GB RAM

---

The latest list of algorithms supported by nx-cugraph can be found in GitHub, or in the Supported Algorithms Section.