|
•
|
The NVIDIA CUDA® direct sparse solver (cuDSS) is an optimized first-generation GPU-accelerated direct sparse solver. This new direct solver can solve some types of problems that require direct solvers significantly faster than traditional CPU-based solvers. It requires installed CUDA driver and runtime versions. The NVIDIA cuDSS direct sparse solver is available for Windows® and Linux®.
|
|
•
|
A new Verlet explicit time-stepping method is now available as an efficient time-stepping method for the new Solid Mechanics, Explicit Dynamics and Truss, Explicit Dynamics interfaces.
|
|
•
|
The Combine Solution feature can now concatenate parametric sweep solutions and convert outer solutions from a parametric sweep to inner solutions.
|
|
•
|
You can perform eigenmode following with respect to a specific parameter in Parametric Sweep and Auxiliary Sweep with an Eigenvalue Parametric solver. A new Mode following checkbox has been added to the Eigenvalue study step and Eigenvalue Solver, and a Mode Following attribute feature has been added.
|
|
•
|
A new option to add an expression-based scaling has been added to the settings for the Eigenvalue Solver. In the Output section, you can now choose Manual from the Scaling of eigenvectors list. An Expression field then becomes available.
|
|
•
|
A new Linearize small mesh elements list in the settings for the Compile Equations node can be used to control the geometric shape-function order of small mesh elements used to compile the equations of the current study. This option only affects the current study and modifies the setting for the Geometry shape function in the Curved Mesh Elements section of the Settings window of the Component node when the Avoid inverted elements by curving interior domain elements option is enabled. The default option is Physics controlled. By specifying the User controlled option, you can enter a length expression for the maximum element size threshold per component in the table below to prevent mesh elements smaller than this threshold from being curved.
|
|
•
|
It is now possible to specify the phase in the Usage section with a stateless interface in reduced-order model features. If not specified, it is by default equal to the phase global variable. In the Parameters table in results features, the name of the phase variable from a reduced-order model like rom1 is rom1.phase; its value is by default given by another variable, rom1.input.phase, which, in turn, is defined by the entry in the new Phase field in the Usage section.
|
|
•
|
A new Reuse null-space data checkbox has been added to the Advanced attribute node. If selected, when solving a model that requires many iterations, the block structure of the matrix N (the constraint matrix) is reused if possible. This speeds up the solver for certain models.
|
|
•
|
All changes to solver settings are now added to the Changes from Default Settings tables, including properties that have become inactive.
|
|
•
|
The MUMPS direct solver has improved support for shared memory parallelism. New Multithreaded matrix factorization and Multithreaded forward and backward solve options have been added to the settings for MUMPS.
|
|
•
|
A new Stop when no eigenvalues are found checkbox has been added to the settings for the Eigenvalue solver to determine whether to return an empty solution or stop the process with an error.
|
|
•
|
You can now specify multiple plots groups to plot while solving in the Results While Solving settings in studies.
|
|
•
|
Improved unit handling for the remeshing time-step setting in Automatic Remeshing nodes. You can now select a Remeshing time step type: Constant or Expression.
|
|
•
|
A Show Location option has been added to the Batch Data node, External Processes node, and External Processes window. It shows the location of the model on the file system or in the version history view in the Model Manager database, depending on where the model is stored.
|
|
•
|
Cluster and batch sweep support has been added for the Surrogate Model Training study node.
|
|
•
|
The Shift used in the modes computation setting has been removed from the LAPACK (filled matrix) eigenvalue solver because it is no longer useful.
|