GPU CUDA Data Precision
In the CUDA API, you can specify the precision with which you want the data to be handled. This allows the API scripts to leverage the speed benefits of the hybrid and single precision solvers.
The three CUDA specific data types are: real, longReal, and customPropReal. The following table shows each data type’s precision depending on the solver settings.
Data Type | Solver Precision | Usage | ||
---|---|---|---|---|
Double | Hybrid | Single | ||
real | double | single | single | Most situations |
longreal | double | double | single | High precision data requirements |
customPropReal | double | double | single | custom properties |
To implement these data types, specify them when declaring a variable as follows:
real exampleVariable = ...
Here, the real data type is used, and is an appropriate data type for most situations. For more information about these data types and the functions associated with them, see the API Help included in the EDEMinstallation folder.