What’s new in Gurobi 14.0?
Gurobi 14.0 delivers major advances in optimization performance:
26% faster MIP performance
>2x faster MINLP performance
23% faster Concurrent LP performance, 24% faster Dual Simplex, 22% faster Barrier
84% faster NL barrier
Better solutions earlier for difficult MIP models
Improved PDHG performance and crossover, expanding the range of LP models that can benefit
New PDHG-based MIP heuristic, with optional GPU acceleration
GPU-enabled machines on Gurobi Instant Cloud
Improved capabilities for nonlinear modeling
Together, these improvements help you solve what was previously out of reach, get better solutions sooner, explore more possibilities, and model more of the real world.
What kind of performance improvements can I expect?
On difficult models taking more than 100 seconds to solve, Gurobi 14.0 delivers approximately 26% faster MIP performance, 23% faster Concurrent LP performance, and over 2x faster global MINLP performance. Individual model performance will vary.
What do these performance improvements mean in practice?
Performance isn’t only about solving the same problem faster. Greater performance can make more demanding problems practical to solve, create room to evaluate more scenarios within the same decision window, and allow you to increase model scope or fidelity without the same increase in solve time.
How does Gurobi 14.0 help me get better solutions sooner?
Gurobi 14.0 significantly improves its ability to find high-quality feasible solutions earlier, particularly for difficult MIP models that may not be practical to solve to proven optimality within the available time. This helps you get more value from the solve time you have.
What’s new with the NoRel heuristic?
Gurobi 14.0 can now run the NoRel heuristic at regular intervals during tree search, rather than only before or instead of the regular search. This increases the solver’s ability to find good solutions earlier on difficult MIP models.
What’s improved for nonlinear optimization?
Gurobi 14.0 delivers over 2x faster MINLP performance, benefiting from improvements across LP relaxations, MIP, nonlinear barrier, convex estimators, and tighter integration across these components. The nonlinear Barrier algorithm has also been improved, delivering 84% faster performance.
What does improved nonlinear performance enable?
Faster nonlinear optimization can make richer mathematical representations practical for more problems, helping you capture nonlinear relationships that might otherwise need to be simplified for computational reasons.
What’s improved for LP?
Gurobi 14.0 delivers approximately 23% faster Concurrent LP performance, primarily through improvements to crossover following Barrier.
Barrier performance improves by approximately 22%, while the performance for Dual Simplex, now multithreaded, also improves by approximately 24%.
What’s improved with PDHG?
PDHG is now used as a heuristic to help find feasible solutions for MIP models, expanding its role in Gurobi 14.0. PDHG performance has improved on both CPUs and GPUs. Gurobi 14.0 can also begin crossover while PDHG is still running, expanding the range of LP models that can benefit from PDHG.
On GPU-enabled machines, PDHG can run on the GPU while crossover runs concurrently on CPU cores.
What is the new IsConvex attribute?
The IsConvex attribute indicates whether Gurobi has determined that a model is convex, ignoring SOS constraints and variable integrality. This makes it easier to assess model structure without parsing Gurobi logs.
Are there any licensing changes in Gurobi 14.0?
Gurobi 14.0 simplifies how concurrency-based license size is measured. Single Use and Floating Use concurrency-based licenses are now expressed in threads, while capacity-based licenses continue to count cores.
