Altair ultraFluidX 2021.2 Release Notes

Highlights

Highlights of this release include:

Seed Runs with EnSight Velocity Fields (Expert Feature): ultraFluidX 2021.2 introduces a basic seeding functionality so ultraFluidX runs can be initialized with previous simulation results. Time-averaged or instantaneous velocity fields can be used. The velocity information is read directly from the standard ultraFluidX EnSight output files. No extra output must be generated to use the feature. If no source information is found (e.g., in case of geometry changes), a global fallback velocity is used to initialize the voxels.
Enhanced Overset Mesh Capabilities (Inclined OSM Zones with Multiple Refinement Levels): OSM regions that rotate around non-axis-aligned axes of rotation are now supported. Previous restrictions to axis-aligned axes of rotation no longer apply.
ultraFluidX 2021.2 can also handle multiple refinement levels in the overset grid regions. The feature must be manually activated in the input deck. By default, OSM regions are automatically refined to the finest RL present in each region. The interface region that connects the overset mesh region to the background mesh must be uniformly refined in any case.

Library Updates (CUDA, Open MPI, and UCX): ultraFluidX 2021.2 utilizes more recent versions of CUDA (version 11.4.0), Open MPI (version 4.1.1) and the Unified Communication X (UCX) framework (version 1.11.0). This does not affect the minimum required CUDA driver version (450.80.02 or higher is required).

A minor inconsistency in the time-averaging has been removed. Now all time-averaged outputs start averaging at the exact same point in time.
Output folders were created by rank 0 of a multi-GPU run. This has been changed to support multi-node runs with local file systems.
The domain size in the effective solver deck did not match the effective domain size of the simulation. Instead, the specified domain size was shown in the effective solver deck, whether it had been corrected or not.
Partial surface values in an Overset Mesh simulation were updated in the coarsest iteration only, even when sub-iteration data was available.
The reported CPU and GPU memory consumption was incorrect and is now correct. The correction increases the CPU memory consumption by around 5% and the GPU memory consumption by around 2.5%.
Sporadically, MRF and OSM zones were refined more than specified or expected.
In isolated cases, voxels outside an MRF and OSM region with geometry intersection were accidentally identified to be inside the zone.