nx_cugraph

nx-cugraph is a NetworkX backend that provides GPU acceleration to many popular NetworkX algorithms.

By simply installing and enabling nx-cugraph, users can see significant speedup on workflows where performance is hindered by the default NetworkX implementation. With nx-cugraph, users can have GPU-based, large-scale performance without changing their familiar and easy-to-use NetworkX code.

Let’s look at some examples of algorithm speedups comparing NetworkX with and without GPU acceleration using nx-cugraph.

Each chart has three measurements.

• NX - default NetworkX, no GPU acceleration

• nx-cugraph - GPU-accelerated NetworkX using nx-cugraph. This involves an internal conversion/transfer of graph data from CPU to GPU memory

• nx-cugraph (preconvert) - GPU-accelerated NetworkX using nx-cugraph with the graph data pre-converted/transferred to GPU

Command line example

Open 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

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:/# NETWORKX_BACKEND_PRIORITY=cugraph ipython bc_demo.ipy

This run will be much faster, typically around 20 seconds depending on your GPU.

user@machine:/# NETWORKX_BACKEND_PRIORITY=cugraph ipython bc_demo.ipy

There is also an option to cache the graph conversion to GPU. This can dramatically improve performance when running multiple algorithms on the same graph.

NETWORKX_BACKEND_PRIORITY=cugraph NETWORKX_CACHE_CONVERTED_GRAPHS=True ipython bc_demo.ipy

When running Python interactively, the cugraph backend can be specified as an argument in the algorithm call.

For example:

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

The latest list of algorithms supported by nx-cugraph can be found here.