Free Courses Fluids Structures 3D Design Electronics Materials Photonics Student Teams STEM Python Missions Optical Installation Indo Kursus My Courses

[FAQ]

1.) All processes step through the code together for each time point. The convergence is a global quantity meaning each processor sums their piece and convergence is decided for the entire model. All processes get the same signal regarding convergence (i.e., converging tells them to all keep going, converged tells them to stop iterating, diverging tells them all to bisect or exit). They all must move together. D-ANSYS can never have some processes stop while others keep going to convergence. 2.) Standard ANSYS and Distributed ANSYS in theory are supposed to return the same exact convergence pattern and same exact answer. However, computer modeling introduces round off errors. Running a program in parallel (shared or distributed memory parallel) introduces different amounts of round off error depending on the number of cores used. For well-conditioned problems this round off error will be insignificant and unnoticeable in the engineering results. However, for unstable problems (or ill-conditioned problems), this round off error can be noticeable and can cause different convergence patterns and different results.