Research map / Learning theory and algorithms / Combinatorial and search optimization
A GPU-based Constraint Programming Solver
Pierre Talbot
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| use ternary constraint network bytecode propagators instead of view-based propagators (ablation) | Turbo GPU solver · MiniZinc Challenge 2024 (20 instances) | nodes per second: 14623 -> 103036 | Improved |
| use TCN representation instead of view-based (ablation) | Turbo GPU solver · MiniZinc Challenge 2024 (20 instances) | propagator memory (MB): 10.33 -> 0.69 | Improved |
| use TCN representation instead of view-based (ablation) | Turbo GPU solver · MiniZinc Challenge 2024 (20 instances) | fixpoint iterations per node: 6.1 -> 13.9 | Hurt |
| add warp-centric fixpoint loop (ablation) | Turbo GPU solver · MiniZinc Challenge 2024 | nodes per second: about +10% | Improved |
| use random split during diving instead of user-defined (ablation) | Turbo dive and solve · MiniZinc Challenge 2024 | instances with idle blocks: 26 -> 10 | Improved |
| use first-fail split during diving instead of user-defined (ablation) | Turbo dive and solve · MiniZinc Challenge 2024 | instances with idle blocks: 26 -> 21 | No clear effect |
| run on H100 instead of Quadro RTX 5000 Max-Q (ablation) | Turbo · MiniZinc Challenge 2024 | average nodes per second: 94k -> 458k | Improved |
| Turbo on GPU vs OR-Tools sequential on CPU | OR-Tools · MiniZinc Challenge 2024 | MiniZinc score: 119.9 -> 45.8 | Hurt |
| Turbo on GPU vs Choco sequential on CPU | Choco · MiniZinc Challenge 2024 | MiniZinc score: 49.3 -> 45.8 | No clear effect |
| Turbo on GPU vs parallel Choco with 64 threads | Choco Par · MiniZinc Challenge 2024 | MiniZinc score: 190.8 -> 45.8 | Hurt |
Closest papers by meaning
- DOGE-Train: Discrete Optimization on GPU with End-to-End Training (AAAI 2024 · similarity 0.89)
- B$^3$-PWL: GPU-Batched Branch-and-Bound for Piecewise-Linear Optimization with SOS2 Constraints (NeurIPS 2026 · similarity 0.87)
- StitchCUDA: An Automated Multi-Agents End-to-End GPU Programing Framework with Rubric-based Agentic Reinforcement Learning (ICML 2026 · similarity 0.84)
- Geometric Algorithms for Neural Combinatorial Optimization with Constraints (NeurIPS 2025 · similarity 0.82)
- BIPNN: Learning to Solve Binary Integer Programming via Hypergraph Neural Networks (NeurIPS 2025 · similarity 0.82)
- Poplar: Efficient Scaling of Distributed DNN Training on Heterogeneous GPU Clusters (AAAI 2025 · similarity 0.82)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
