Solving Larger Models in COMSOL Multiphysics^®

Name: Solving Larger Models in COMSOL Multiphysics®
Uploaded: 2021-01-11 20:16:00
Duration: 1 h 4 min 25 s
Description: Get an introduction to high-performance computing and solving large models with COMSOL Multiphysics®. Watch this archived webinar.

Durée: 1:04:25

Retour aux videos

Aperçu
Transcription

Watch this archived webinar for an introduction to high-performance computing with COMSOL Multiphysics^®. Both algorithmic and hardware aspects of multicore and cluster computing are discussed.

You will also learn about the significant performance improvements for the domain decomposition and algebraic multigrid solver technology in COMSOL Multiphysics^® available as of version 5.6. These solver technologies have been extended with new algorithms for both CFD and acoustics.

Browse upcoming live webinars here.

Chapter Selection

Introduction (0:00) Outline (0:23) Solvers (2:58) Predicting Solver Memory Requirements (4:26) Five Test Models (5:48) DOF Limits (6:40) Matrix Characteristics (6:59) How Many DOFs in My Model? (8:18) Different Elements and Orders (8:51) Understanding Paralellism (11:22) A CPU with Multiple Cores (12:28) Benefit of Multicore (14:15) Serial and Parallel Work (16:17) Helical Static Mixer (17:44) Increasing Size (19:48) Significant Memory and CPU Requirements (20:38) Out of Memory (21:17) Submodeling (24:14) Coarsening of the Discretization (25:07) Direct Solvers Can Go “Out of Core” (27:59) Temporary Files (28:32) Linear System Solvers (28:43) Complexity of Direct Methods (29:44) Direct and Iterative Solvers (30:20) Let’s Compare Direct with Iterative (31:12) Reducing Memory Usage (33:30) Disable Reuse (34:11) Reduce Number of Cores (34:18) Assembly Settings Affecting Memory (34:28) Ways to Get More Memory (35:09) An Introduction to Clusters (37:13) Using a Cluster with COMSOL Multiphysics^® (37:36) Passing Data Between Nodes of Cluster Can Get Very Complicated (38:38) Distributed Computing with COMSOL Multiphysics^® (40:51) Ethernet Vs. Infiniband (41:04) Example 1: Stationary CFD Using a Cluster (41:43) Benchmark: Airflow Over an Ahmed Body (43:20) Multigrid CPU Time (43:41) Multigrid Memory (44:22) DD Vs. MG CPU Time (44:45) DD Vs. MG Memory (45:45) Example 2: Stationary Semiconductor Models Using a Cluster (46:56) Benchmark: bipolar_transistor_3d (2M DOF) (48:23) Benchmark: bipolar_transistor_3d (49:31) Speedup and Memory Gain for bipolar_transistor_3d (50:17) Parameter Variation (52:46) Only Storing Parts of the Solution (53:24) Probing and Saving on File (54:12) Summary (55:03) Q&A (57:11)