Release 2024 R2 includes the following improvements to the solution process:
- 3.1. Distributed-Memory Parallel Processing Enhancements
- 3.2. Enhanced Memory Reduction for Models with Constraint Equations and High Core Counts
- 3.3. Reduced Peak Memory Usage for Uncompressed Node and Element Numbering
- 3.4. Support for Devices for the GPU Accelerator
- 3.5. Improved Performance for the PCG Solver
- 3.6. Support for Mixed u-P Formulation with the PCG Solver
- 3.7. Beta Feature - Frequency Domain Decomposition for Modal Analyses
Enhancements include:
Modifications to the default processor pinning strategy and communication protocols (supplied to Open MPI as command-line options) yielded competitive performance on both single machines and HPC clusters with AMD-based processors.
The default support for message passing interface (MPI) software has been upgraded to Intel MPI 2021.11.0 for both Linux and Windows. In prior releases, the default was Intel MPI 2021.10.0. For details and other optional supported MPI software versions, see Platforms and MPI Software in the Parallel Processing Guide
The modified architecture of constraint equations significantly reduces memory for models containing a large number of constraint equations and using high core counts (32 or more cores).
The internal data structures used to store number mappings have been redesigned, with a particular focus on both node and element number mappings. This resulted in a significant reduction of Mechanical APDL peak memory usage when the numbering is highly uncompressed, for example when the maximum defined node or element number is many times greater than the number of defined nodes/elements.
For Linux and Windows, the following cards have been tested and added to the list of recommended GPU devices for the GPU accelerator:
NVIDIA L40
NVIDIA RTX 6000 Ada
NVIDIA RTX 5000 Ada
NVIDIA RTX A5500
NVIDIA RTX A4500
For the complete list of recommended GPU accelerators, see Requirements for the GPU Accelerator in Mechanical APDL (Linux) and Requirements for the GPU Accelerator in Mechanical APDL (Windows) in the Ansys Installation Guides.
The performance of the PCG solver when using one or more GPUs has been improved. Starting in this release, unsymmetric matrices are supported with the PCG solver when using one or more GPUs. Also, when using the PCG Lanczos eigensolver, additional calculations have been offloaded to the GPU for acceleration, which further improves the performance.
For static analyses and transient analyses that use the full method, the PCG
solver now supports the mixed u-P element formulation option for the following
element types: PLANE182, PLANE183,
MPC184, SOLID185,
SOLID186, and SOLID187. This
is the new default for the LM_KEY option of
PCGOPT (LM_KEY = OFF).
Modal analyses now support the use of frequency domain decomposition (DDOPTION,FREQ) as a beta feature. For details, see Frequency Domain Decomposition for Modal Analyses in the Mechanical APDL Beta Features.