- Info
Session 6
-
MPP Execution of Implicit Mechanics with 10M or More Elements
-
LS-DYNA models for Implicit Mechanics are getting larger and more complex. We are continually
seeing models where the linear algebra problems in Implicit Mechanics have 3 to 5 million elements
and know of at least one that is nearly 30M elements. It is these very large linear algebra problems
that distinguish the computer requirements for Implicit Mechanics. This paper will present a study of
the performance of the MPP implementation of implicit mechanics in LS-DYNA examining such issues
as performance, speed-up, and requirements for computer configuration.
-
Progress on GPU Implementation for LS-DYNA Implicit Mechanics
-
Graphics processing units (GPUs) are ubiquitous devices designed to improve the end-user
experience in mass market arenas such as gaming. High-end GPUs have an order of magnitude
more computing power than their hosts, and are thus attractive candidates for accelerating compute
bound applications such as MCAE. This talk will present how we have extended LS-DYNA to utilize
Nvidia Tesla GPUs for implicit mechanics. We will describe the target environment along with
performance results on a range of benchmark problems. The performance results will illustrate when
it makes sense today to utilize the GPU, and when it does not.
-
Performance Benefits of NVIDIA GPUs for LS-DYNA
-
This work examines the performance characteristics of LS-DYNA for the latest CPU and GPU
technologies available. The results are provided for system configurations of workstations for finite
element models that are relevant to current industry practice. The motivation for these studies was to
quantify the parallel performance benefits of LS-DYNA for the latest generation GPU from NVIDIA, the
Tesla 20-series (codenamed Fermi) for implicit finite element models with 100K and greater DOFs,
and for static and dynamic response load conditions.
-
Performance of Large Scale Implicit Crash Analysis on Multicore Processor Systems
|
