For the GPU solving the installation of CUDA Driver is necessary, more information here.
To run the optimization an optimisation on one or more graphics card(s) , Cuda Driver needs to be installed once (further information for Windows and Linux) . After this the solver needs to be started as follows (for Windows: use a Powershell):
...
mscgd2019_CUDASolver 1234
...
The IP and port must be specified in the configuration file "FEM" (Standard "localhost" and "1234").
...
After activating the Cuda Solver, the optimization can be started.
...
The solver can be used by several optimizations and doesn't have to be started again for following optimizations.
...
the solver priority can be set to GPU solving in the Application Settings - Generative Design Solver. The default solver setting is CPU. GPU solving should be preferred because of increased solving speed with the associated license.
Prior to the GPU solving the CPU threads should be defined because they are used to build up the FE-matrix and the set value is important for CPU as well as GPU computing. Afterwards the GPU solving can be activated and the number of GPUs can be added.
Further information here.