You can override the precision set on the Run Definition tab by selecting Override Default Precision and then setting the precision. For details on the precision of executables, see Double-Precision Executables.

You can override the problem size capability ("large problem" or not) set on the Run Definition tab by selecting Override Default Large Problem Setting and then setting the problem size capability (via the Large Problem check box). For details on the problem size capability of the executables, see Large Problem Executables.

Under Partitioning Detail, you can specify various partition method options.

Partition smoothing attempts to minimize the surface area of partition boundaries by swapping vertices between partitions. Partition smoothing reduces communication overhead and improves solver robustness. Smoothing is enabled by default, but may be disabled by changing Partition Smoothing > Option.

If smoothing is enabled, the algorithm will, by default, perform a maximum of 100 sweeps. The maximum number of smoothing sweeps can be specified by changing the value of Partition Smoothing > Max. Smooth. Sweeps. The smoothing algorithm will stop before this value is reached if it finds no improvement between successive sweeps.

For further details on partition smoothing, see Optimizing Mesh Partitioning in the CFX-Solver Modeling Guide.

The partitioning in CFX is node based; the partitioner distributes the nodes among the partitions. By default, the distribution does not consider the computational cost per node as a function of mesh element type (for example, tetrahedral or hexahedral), leading to potential load imbalances between different partitions; for example, two partitions with the same number of nodes could have different computational loads if they have different proportions of tetrahedral (more expensive) versus hexahedral (less expensive) elements. Such load imbalances between the partitions can greatly reduce parallel performance. In order to avoid or reduce such load imbalances – especially for meshes with mixed element types – you can apply a Partition Node Weighting option to influence the node counts among the partitions:

This option automatically creates a coarse version of the mesh before partitioning that into the desired number of partitions. The coarsening process is then reversed to give the partitioning of the original mesh. This additional pre-coarsening step uses the same AMG technology as the solver, and allows the partitioning to be more 'aware' of the solution process. The method can be used to prevent partition boundaries from passing through areas of high aspect ratio cells, and can resolve convergence difficulties that this can cause, particularly with diffusion-only type equations.

There are two sub-options:

The Mesh Pre-coarsening option is compatible with all the basic partitioning methods, such as MeTiS, and Optimised Recursive Coordinate Bisection, but will have most impact on methods that are not purely geometrical and involve a connectivity graph. Note that the option carries significant overhead at partition-time, in terms of both memory and CPU usage.

If required, you can adjust the memory configuration under Partitioner Memory. For details, see Configuring Memory for the CFX-Solver in the CFX-Solver Manager User's Guide.