About Solver

The kernel steps of the solver can be launched on a different process.

The below capability is useful only for Windows 32-bit platforms. The kernels steps of the solver are launched transparently on a different process. It concerns the steps that are consuming a lot of memory. The slave process will benefit from small contiguous available memory for computation.

It is strongly recommended that you extend the memory of the used machine with extended paginated memory. The master process will automatically paginate its own data on this paging memory.

You can improve performance on multi-processor computers.

The computation process is divided in two steps.

The computation of an object requires two distinct actions:

• First, the user-defined specifications attached to each object belonging to the objects set in the specification tree must be translated by the pre-processor into solver-interpretable commands. Since solvers can only interpret data applied on mesh entities (nodes and elements), this first translation step requires the existence of a mesh support for converting user input specifications on the geometry into explicit solver commands on nodes and elements.
• Next, the solver translates the solver commands into data ready for algorithmic treatment as required by the numeric procedure invoked. Since algorithms deal only with operators dimensioned by the problem size (number of degrees of freedom), this second translation step requires the exact knowledge of the behavioral hypotheses of the finite elements (properties), which contain the required degree of freedom information.

The program translates the user-defined specifications into solver-interpretable commands applied on mesh entities, and you can visualize on the mesh the result of this translation.