Benchmarks
Performance measurements across LP, QP, and MILP problem instances.
Important Disclaimer
Most benchmark data is pending. Results will be updated as the solver matures.
Instance | Class | Variables | Constraints | Non-zeros | Runtime | Objective | Gap | Status | Hardware |
|---|---|---|---|---|---|---|---|---|---|
blend.mps | LP | 114 | 74 | 521 | 12.5 ms | -3.0812e+1 | 1.00e-6 | Measured | Intel Core i7-12700H |
ieee118.mps | LP | 236 | 118 | 854 | 45.2 ms | 1.2966e+5 | 1.00e-5 | Measured | Intel Core i7-12700H |
ieee118.mps | LP | 236 | 118 | 854 | Pending | - | - | Pending | NVIDIA RTX 3070 Ti |
prodplan.mps | MILP | 850 | 420 | 2,840 | Pending | - | - | Pending | Intel Core i7-12700H |
Benchmark Methodology
Our benchmarks are designed to provide a realistic assessment of the solver's capabilities across different problem classes.
- Hardware: Tests are run across CPU and CUDA backends on standard consumer and workstation hardware.
- Problem Instances: We use industry-standard libraries including Netlib for LP, MIPLIB for MILP, QPLIB for QP, and synthetic industrial instances.
- Validation: Solutions are strictly validated for KKT conditions and primal-dual feasibility.
Use the Solver Studio to run your own benchmarks interactively.