Benchmarks#
This page presents the performance of algorithms currently supported by nx-cugraph across four graphs of varying sizes.
The goal is to provide a clear comparison of how dispatching to a GPU-accelerated backend compares against the default, CPU-based implementation of NetworkX. This allows users to get an idea of potential speedups that they might see by leveraging dispatching to a GPU with nx-cugraph.
As datasets grow larger, the GPU-accelerated backend typically begins to show increasingly faster speedups over the CPU. This trend demonstrates how the GPU’s parallel processing capabilities allow it to handle large-scale graph analytics much more efficiently than the CPU.
Datasets#
The datasets used in the benchmarks are:
| Dataset | Number of Nodes | Number of Edges | Description |
|---|---|---|---|
| netscience | 1,461 | 5,484 | A small, directed graph representing a scientific collaboration network. |
| amazon0302 | 262,111 | 1,234,877 | A medium, undirected graph modeling product co-purchasing data. |
| cit-Patents | 3,774,768 | 16,518,948 | A large, directed graph representing U.S. patent citation relationships. |
| soc-LiveJournal1 | 4,847,571 | 68,993,773 | A very large, directed graph based on the LiveJournal social network. |
Algorithm Table#
The table below showcases the performance difference between CPU and GPU when running a particular algorithm on one of the four datasets.
Each cell displays the speedup factor achieved when running the algorithm on a nx-cugraph compared to default NetworkX. A higher value indicates better performance. To enhance readability, the cells are color-coded based on the following speedup criteria:
Green for excellent (
>5.0)Yellow for moderate (
1.1 - 5.0)Red for suboptimal (
0 - 1.0)
Also, by hovering over the table, users can see the time it took for NetworkX and nx-cugraph respectively, as well as the delta between both runs. We encourage users to refer to this table to see an approximation of the performance increase they can expect from their workflows.
| Algorithm | netscience | amazon0302 | citpatents | livejournal |
|---|---|---|---|---|
| ancestors | 0.089*nx: 491.479us nxcg: 5.497ms delta: -5.005ms |
60.463nx: 1.712s nxcg: 28.310ms delta: 1.683s |
0.134*nx: 4.359ms nxcg: 32.491ms delta: -28.132ms |
78.609nx: 56.688s nxcg: 721.145ms delta: 55.967s |
| average_clustering | 1.619*nx: 14.918ms nxcg: 9.213ms delta: 5.706ms |
293.432nx: 6.745s nxcg: 22.988ms delta: 6.722s |
868.635nx: 181.956s nxcg: 209.474ms delta: 181.747s |
2828.128nx: 2151.410s nxcg: 760.719ms delta: 2150.649s |
| betweenness_centrality, k=10 | 0.274*nx: 5.449ms nxcg: 19.923ms delta: -14.474ms |
56.98nx: 14.256s nxcg: 250.186ms delta: 14.005s |
99.364nx: 95.752s nxcg: 963.653ms delta: 94.789s |
264.882nx: 610.395s nxcg: 2.304s delta: 608.091s |
| betweenness_centrality, k=100 | 0.239*nx: 42.450ms nxcg: 177.676ms delta: -135.225ms |
50.352nx: 110.614s nxcg: 2.197s delta: 108.417s |
520.555nx: 966.550s nxcg: 1.857s delta: 964.694s |
487.469nx: 6688.933s nxcg: 13.722s delta: 6675.211s |
| bfs_edges | 0.035*nx: 295.556us nxcg: 8.488ms delta: -8.192ms |
0.383nx: 1.260s nxcg: 3.292s delta: -2.032s |
0.003*nx: 115.556us nxcg: 35.910ms delta: -35.794ms |
2.141nx: 75.646s nxcg: 35.329s delta: 40.318s |
| bfs_layers | 0.046*nx: 307.147us nxcg: 6.690ms delta: -6.383ms |
9.489*nx: 327.838ms nxcg: 34.548ms delta: 293.290ms |
0.004*nx: 122.647us nxcg: 32.934ms delta: -32.811ms |
33.468nx: 16.313s nxcg: 487.419ms delta: 15.825s |
| bfs_predecessors | 0.04*nx: 336.403us nxcg: 8.442ms delta: -8.106ms |
0.389nx: 1.282s nxcg: 3.293s delta: -2.011s |
0.004*nx: 139.603us nxcg: 35.985ms delta: -35.845ms |
1.728nx: 61.628s nxcg: 35.664s delta: 25.964s |
| bfs_successors | 0.039*nx: 329.301us nxcg: 8.441ms delta: -8.112ms |
0.491nx: 1.802s nxcg: 3.669s delta: -1.866s |
0.004*nx: 130.468us nxcg: 35.864ms delta: -35.733ms |
1.822nx: 62.495s nxcg: 34.297s delta: 28.198s |
| bfs_tree | 0.121*nx: 855.597us nxcg: 7.060ms delta: -6.204ms |
58.569nx: 2.929s nxcg: 50.011ms delta: 2.879s |
0.011*nx: 403.871us nxcg: 35.151ms delta: -34.748ms |
55.688nx: 74.665s nxcg: 1.341s delta: 73.324s |
| clustering | 1.596*nx: 14.618ms nxcg: 9.162ms delta: 5.456ms |
136.185nx: 6.775s nxcg: 49.747ms delta: 6.725s |
168.938nx: 181.769s nxcg: 1.076s delta: 180.693s |
1132.732nx: 2149.226s nxcg: 1.897s delta: 2147.328s |
| core_number | 0.223*nx: 2.372ms nxcg: 10.652ms delta: -8.280ms |
22.456nx: 1.520s nxcg: 67.706ms delta: 1.453s |
-- | -- |
| degree_centrality | 0.331nx: 361.917us nxcg: 1.092ms delta: -730.266us |
2.834*nx: 75.040ms nxcg: 26.474ms delta: 48.566ms |
2.037*nx: 1.565s nxcg: 768.326ms delta: 796.394ms |
2.045nx: 2.048s nxcg: 1.002s delta: 1.047s |
| descendants | 0.059*nx: 325.657us nxcg: 5.552ms delta: -5.226ms |
38.477nx: 1.259s nxcg: 32.708ms delta: 1.226s |
0.004*nx: 127.983us nxcg: 31.525ms delta: -31.397ms |
82.354nx: 61.836s nxcg: 750.849ms delta: 61.085s |
| descendants_at_distance | 0.007*nx: 26.257us nxcg: 3.593ms delta: -3.567ms |
0.003*nx: 20.914us nxcg: 6.145ms delta: -6.124ms |
0.001*nx: 32.306us nxcg: 30.265ms delta: -30.233ms |
0.065*nx: 5.894ms nxcg: 90.867ms delta: -84.973ms |
| edge_betweenness_centrality, k=10 | 0.167*nx: 7.114ms nxcg: 42.690ms delta: -35.576ms |
14.605nx: 18.773s nxcg: 1.285s delta: 17.488s |
10.153nx: 121.627s nxcg: 11.979s delta: 109.647s |
16.778nx: 909.093s nxcg: 54.182s delta: 854.911s |
| edge_betweenness_centrality, k=100 | 0.15*nx: 43.901ms nxcg: 292.226ms delta: -248.325ms |
24.172nx: 172.498s nxcg: 7.136s delta: 165.361s |
68.332nx: 972.557s nxcg: 14.233s delta: 958.325s |
112.81nx: 8546.360s nxcg: 75.759s delta: 8470.601s |
| ego_graph | 0.726*nx: 5.584ms nxcg: 7.693ms delta: -2.109ms |
104.093nx: 5.857s nxcg: 56.265ms delta: 5.801s |
0.008*nx: 1.431ms nxcg: 187.666ms delta: -186.235ms |
150.036nx: 302.519s nxcg: 2.016s delta: 300.503s |
| eigenvector_centrality | 7.409*nx: 28.642ms nxcg: 3.866ms delta: 24.776ms |
-- | 263.997nx: 242.879s nxcg: 920.007ms delta: 241.959s |
376.161nx: 458.153s nxcg: 1.218s delta: 456.935s |
| generic_bfs_edges | 0.032*nx: 276.460us nxcg: 8.570ms delta: -8.294ms |
0.38nx: 1.238s nxcg: 3.256s delta: -2.018s |
0.003*nx: 109.543us nxcg: 35.921ms delta: -35.811ms |
2.129nx: 75.324s nxcg: 35.374s delta: 39.949s |
| hits | 1.703*nx: 7.290ms nxcg: 4.280ms delta: 3.010ms |
-- | 33.932nx: 64.038s nxcg: 1.887s delta: 62.150s |
87.124nx: 233.893s nxcg: 2.685s delta: 231.209s |
| in_degree_centrality | 0.459nx: 293.520us nxcg: 639.090us delta: -345.570us |
2.095*nx: 57.231ms nxcg: 27.314ms delta: 29.917ms |
1.645*nx: 1.275s nxcg: 775.082ms delta: 499.672ms |
1.808*nx: 1.798s nxcg: 994.536ms delta: 803.470ms |
| is_weakly_connected | 0.001*nx: 28.767us nxcg: 31.511ms delta: -31.482ms |
9.442*nx: 469.725ms nxcg: 49.747ms delta: 419.978ms |
52.655nx: 10.313s nxcg: 195.850ms delta: 10.117s |
64.03nx: 22.803s nxcg: 356.123ms delta: 22.446s |
| k_truss | 8.31*nx: 12.342ms nxcg: 1.485ms delta: 10.857ms |
539.652nx: 6.103s nxcg: 11.309ms delta: 6.092s |
-- | -- |
| katz_centrality | 122.153nx: 4.219s nxcg: 34.535ms delta: 4.184s |
648.055nx: 24.870s nxcg: 38.376ms delta: 24.831s |
300.147nx: 274.585s nxcg: 914.835ms delta: 273.670s |
24767.824nx: 76304.239s nxcg: 3.081s delta: 76301.159s |
| louvain_communities | 2.479*nx: 72.361ms nxcg: 29.185ms delta: 43.176ms |
103.205nx: 43.659s nxcg: 423.034ms delta: 43.236s |
272.652nx: 2409.300s nxcg: 8.837s delta: 2400.464s |
208.843nx: 4291.899s nxcg: 20.551s delta: 4271.348s |
| number_weakly_connected_components | 0.041*nx: 1.296ms nxcg: 31.745ms delta: -30.449ms |
9.956*nx: 493.798ms nxcg: 49.599ms delta: 444.199ms |
54.189nx: 10.795s nxcg: 199.216ms delta: 10.596s |
64.136nx: 23.510s nxcg: 366.558ms delta: 23.143s |
| out_degree_centrality | 0.453nx: 287.390us nxcg: 634.436us delta: -347.046us |
2.156*nx: 58.691ms nxcg: 27.227ms delta: 31.464ms |
1.549*nx: 1.202s nxcg: 775.760ms delta: 426.122ms |
1.727*nx: 1.714s nxcg: 992.363ms delta: 721.574ms |
| overall_reciprocity | 7.901*nx: 15.359ms nxcg: 1.944ms delta: 13.415ms |
563.312nx: 4.525s nxcg: 8.032ms delta: 4.517s |
1524.889nx: 78.576s nxcg: 51.529ms delta: 78.524s |
2257.021nx: 239.907s nxcg: 106.294ms delta: 239.801s |
| pagerank | 1.696*nx: 6.589ms nxcg: 3.885ms delta: 2.704ms |
76.288nx: 2.828s nxcg: 37.065ms delta: 2.791s |
82.128nx: 72.134s nxcg: 878.314ms delta: 71.256s |
188.302nx: 223.766s nxcg: 1.188s delta: 222.578s |
| pagerank_personalized | 1.406*nx: 6.942ms nxcg: 4.939ms delta: 2.003ms |
55.666nx: 3.265s nxcg: 58.648ms delta: 3.206s |
46.6nx: 65.932s nxcg: 1.415s delta: 64.517s |
120.002nx: 237.002s nxcg: 1.975s delta: 235.027s |
| reciprocity | 0.01*nx: 22.615us nxcg: 2.313ms delta: -2.291ms |
0.013*nx: 43.303us nxcg: 3.431ms delta: -3.388ms |
0.009*nx: 54.886us nxcg: 5.864ms delta: -5.809ms |
0.172*nx: 2.705ms nxcg: 15.737ms delta: -13.032ms |
| shortest_path | 0.064*nx: 393.284us nxcg: 6.105ms delta: -5.711ms |
1.1*nx: 960.214ms nxcg: 872.857ms delta: 87.358ms |
0.005*nx: 168.884us nxcg: 33.088ms delta: -32.919ms |
1.344nx: 49.690s nxcg: 36.964s delta: 12.726s |
| single_source_shortest_path_length | 0.042*nx: 235.942us nxcg: 5.613ms delta: -5.377ms |
6.285*nx: 280.148ms nxcg: 44.572ms delta: 235.576ms |
0.003*nx: 89.561us nxcg: 31.926ms delta: -31.836ms |
14.241nx: 15.347s nxcg: 1.078s delta: 14.269s |
| single_target_shortest_path_length | 0.042*nx: 240.253us nxcg: 5.682ms delta: -5.442ms |
5.482*nx: 290.073ms nxcg: 52.910ms delta: 237.162ms |
0.098*nx: 3.321ms nxcg: 33.767ms delta: -30.446ms |
7.672nx: 11.358s nxcg: 1.480s delta: 9.878s |
| transitivity | 1.682*nx: 15.338ms nxcg: 9.116ms delta: 6.222ms |
331.111nx: 7.525s nxcg: 22.726ms delta: 7.502s |
937.121nx: 196.579s nxcg: 209.769ms delta: 196.369s |
2831.727nx: 2157.310s nxcg: 761.836ms delta: 2156.548s |
| triangles | 0.625*nx: 5.069ms nxcg: 8.105ms delta: -3.036ms |
56.354nx: 2.591s nxcg: 45.977ms delta: 2.545s |
55.831nx: 55.304s nxcg: 990.560ms delta: 54.313s |
124.634nx: 220.697s nxcg: 1.771s delta: 218.926s |
| weakly_connected_components | 0.035*nx: 1.275ms nxcg: 36.601ms delta: -35.326ms |
7.677*nx: 493.186ms nxcg: 64.240ms delta: 428.946ms |
11.559nx: 10.928s nxcg: 945.427ms delta: 9.983s |
20.597nx: 24.027s nxcg: 1.167s delta: 22.861s |
* indicates that the time difference between CPU and GPU was insignificant (<1s)