ANSYS, Inc. Release Notes
ANSYS, Inc. Release Notes
Release 14.0 includes the following new enhancements that improve solution procedures and features.
The following enhancements are available for Distributed ANSYS:
Support for GPU acceleration has been added. See GPU Acceleration Enhancements for more details.
You can now avoid combining the local or distributed results files into a single, global results file upon completion of the solution. The file-combination control is also available for other solution files. See the DMPOPTION and RESCOMBINE commands for more information.
Support for the new subspace iteration (SUBSP) eigensolver (for eigenvalue buckling analyses only) is available. See the BUCOPT command for more information.
Support for TRANS126, INFIN110, INFIN111, PLANE121, and PLANE230 element types has been added.
Support is available for the EFLG option on the NLDIAG command.
Analyses involving contact elements are much more robust when restarting the analysis (that is, when performing a multiframe restart).
New error-handling logic has been added to avoid deadlocks (hung jobs) if any unexpected error occurs during the parallel job execution. If such an error occurs, diagnostic information is now printed into one of the output files written by each Distributed ANSYS process.
Radiosity surface elements SURF251 and SURF252 are now supported.
The following enhancements are available for the GPU Accelerator capability.
Support for Distributed ANSYS.
Support includes both multicore servers and clusters (that is, single-machine and multiple-machine hardware). In this release, only one GPU per machine or computing node is supported. For example, when using Distributed ANSYS on a cluster involving eight computing nodes with each computing node having two GPUs, only a single GPU per node (a total of eight GPUs) can be used to accelerate the simulation.
Support for the NVIDIA Quadro 6000 card.
Support for modal analyses using the unsymmetric or damped eigensolver (MODOPT,UNSYM or MODOPT,DAMP).
Improved performance relative to the previous release.
When using the sparse solver, the solver kernel running on the GPU hardware is up to 25 percent faster than the prior release. When using the PCG/JCG solvers, the solver kernel that is run on the GPU hardware is up to 40 percent faster than the prior release.
A new subspace eigensolver (BUCOPT,SUBSP) is available for eigenvalue buckling analyses. The eigensolver uses essentially the same algorithm as the unsymmetric eigensolver (MODOPT,UNSYM) to solve the generalized eigenvalue problem.
The subspace eigensolver is most appropriate for linear perturbation eigenvalue buckling analyses in which the tangent stiffness matrix becomes indefinite. In such cases, the subspace eigensolver is more likely to achieve a successful solution compared to the Block Lanczos eigensolver.
Overconstraint detection is now available and includes the topological method and the algebraic method. In the algebraic method, the constraint equations introduced by the CE and CP commands, and by P (pressure) variables from the element u-P formulation, are taken into account.
The following are solver-related changes and enhancements.
The performance of the sparse solvers (both shared memory and distributed memory; EQSLV,SPARSE) has been enhanced when running on "AVX SIMD" capable Intel and AMD processors (for example, Intel Xeon processors code-named "Sandy Bridge"). In some cases, the solver performance can be up to 50 percent faster than the previous release when running on this specific hardware.
The performance of the multiframe restart procedure has been greatly improved, particularly when many boundary condition specifications exist (D, F, CE, etc.) or when many load steps are involved. In some cases, the performance of the restart action is now five times faster than the previous release.
The performance of the shared memory sparse solver (EQSLV,SPARSE) has been enhanced. In some cases, the solver performance can be up to 40 percent faster than the previous release, regardless of the processor hardware used.
The PCG solver now supports the Lagrange multiplier
method of the MPC184 family of elements.
The imposed Lagrange multipliers are transferred into multiple point
constraints so that the PCG solver can be used to obtain a solution.
To activate this functionality, the LM_Key field on the PCGOPT command must be set to ON.