A large number of solvers for mathematical programming models have been hooked up to GAMS. The tables below provide a brief description of each solver, the model types each solver is cabable of solving, and the platforms supported by each solver. For general information on using GAMS solvers, see Solver Usage.

Model Types

GAMS is able to formulate models in many different types of problem classes or model types. Typically, a solver will be capable of solving (i.e. will accept as input) more than one model type. The solver/model type matrix shows which solver is capable of which model type:

* deterministic global solver ** Here model type EMP excludes EMP-SP

When choosing a solver, some judgement should be applied when considering the listed model type capabilities for the solver - the same capability "check boxes" does not imply equality in capacity or suitability. For example, take a hypothetical solver WeOpt designed to solve MINLP models. Since the problem class MINLP includes NLP, MIP, and LP as subclasses, solver WeOpt could include these capabilities also. If WeOpt is also a good performer on NLP models, it would include that capability. But if it does not shine at all as a MIP or LP solver, we would choose not to include MIP and LP in the capability list for WeOpt. In such a case one can always solve using a more general model type (e.g. solve an LP model as NLP so WeOpt can be used), but WeOpt will not advertise itself as an LP solver. Since the WeOpt solver does not even recognize MCP or MPEC models, we don't include those capabilities.

There are two types of non-linear solvers: local and global. A local solver can find a local optimum but generally cannot comment on global optimality of the solution. A local optimum is a point where the objective value is better than its neighboring points but could be worse than a distant point. On the other hand, a global solver can find and prove that the obtained solution is a global optimum i.e., there is no feasible solution that would result in a better objective value. It is worth noting that a local solver can solve a model to global optimality in some special cases. In the table shown, the entries with * in the column Global indicate solvers that can solve a model to global optimality.

There are other differences in solvers that are difficult to quantify or cannot be captured by a capability table like the one shown. For example, for nonconvex NLP or QCP models, one solver could look only for first-order stationary points, another for local solutions, a third for local solutions using a scatter search or similar search heuristic, and a fourth could do a true global search for the global optimum. The relative merits (measured typically by speed alone) of solvers is the subject of considerable benchmarking activity and discussion.

The GAMS sales team can help answer questions you may have about solver capability. We also offer free evaluation licenses to help you decide what solvers are most suitable for your models.

Supported Platforms

The solver/platform matrix shows which platforms each solver is supported on. In addition, where a vendor has discontinued solver support for a particular platform and we continue to ship the last available supported version, this version number is indicated as well.