Hi all,
I wanted to share GravOpt, a MAX-CUT solver optimized for large graphs on a single CPU core. It’s particularly useful for graph optimization tasks in ML / AI pipelines, where GPU is not available or CPU-only solutions are preferred.
Open-Source Demo
-
GravOpt Adaptive E – ★5, already tested on 20k-node graphs
-
Performance (demo limits):
-
20k nodes → ~7 min on a single CPU core
-
Demo limited to 20 nodes for testing
-
GravOpt Pro – Unlimited Version
-
Tested on up to 1.2M nodes
-
Unlimited nodes & iterations
-
Full source code
-
All future updates included
-
Priority email support
Purchase & details: GravOpt Pro
Benchmark vs common Python solvers (CPU-only)
| Solver | Nodes | Time (min) | Notes |
|---|---|---|---|
| GravOpt Adaptive E (demo) | 20,000 | 7 | Demo limit 20 nodes |
| GravOpt Pro | 1,200,000 | 180 | Unlimited, tested |
| NetworkX max_cut | 20,000 | 45+ | CPU-only |
| PyG / Torch MAX-CUT | 20,000 | 30+ | GPU recommended |
GravOpt Pro scales efficiently on large graphs without GPU.
Python Integration Example
import networkx as nx
from gravopt import gravopt_maxcut
# Create a sample graph
G = nx.erdos_renyi_graph(5000, 0.01)
# Run MAX-CUT optimization
cut_value, partition = gravopt_maxcut(G, iterations=1000)
print(f"Max-CUT value: {cut_value}")
-
Works directly with NetworkX graphs
-
Easy to integrate into PyTorch or NumPy pipelines
If you are experimenting with large-scale graph optimization or AI heuristics, GravOpt can be a powerful CPU-only alternative.
Happy to discuss benchmarks, scaling, or integration.
Dimitar Kretski
GitHub GitHub - Kretski/GravOpt-Pro: GravOpt Pro – Unlimited MAX-CUT Solver · €200 lifetime license (first 100)