Computational Electromagnetics

Overview

Computational ElectroMagnetics (CEM) first became a significant HPC application during the development of stealth aircraft during the 1980s and remains an important application for many of today’s most demanding HPC users. These methods model the interactions of electromagnetic fields with physical objects and the environment. CEM methods are used for numerous applications including:

Boundary Elements Methods

There are several methods for modeling electromagnetic systems, including Boundary Element Methods (BEM), sometimes also known as the Method of Moments (MoM). BEM methods are more computationally expensive than some of the alternatives, such as Fast Multipole Methods (FMM), but BEM techniques have several advantages that keep them in popular use today.For example, a BEM approach may yield the right solution in situations where an FMM approach fails to converge.

Accelerated Boundary Elements Methods

ClearSpeed’s accelerators are ideal for speeding up BEM applications where most of the computational cost is in complex L3 BLAS functions such as ZGEMM, of which ClearSpeed provides accelerated versions in its CSXL math library. This makes it possible to transparently accelerate a BEM application by linking against CSXL in addition to the existing L3 BLAS libraries being used.

Real-World Acceleration Example

Multipath is a leading provider of out-of-core linear solver libraries. Multipath’s library, called FMSLIB, is used for many applications, including finding solutions to BEM problems. Multipath has been working with ClearSpeed to accelerate FMSLIB. By accelerating the ZGEMM L3 BLAS function called by FMSLIB, ClearSpeed has shown a3-5X speedup using its Advance accelerator boards, depending on the host system.

Benefits

BEM users are limited by the length of time their simulations take. Now ClearSpeed makes it possible to significantly increase your system’s BEM performance, enabling more accurate simulations on bigger models than ever before, or reducing turnaround time on models of a given size. ClearSpeed’s CSXL math library uses transparent heterogeneous processing techniques to aggregate the performance of the accelerator with the multi-core host. This ensures that ClearSpeed’s accelerators always add performance to your system, and with the range of accelerator products now available from ClearSpeed, any system, from 1U servers through to blades can receive that acceleration benefit. ClearSpeed’s unique enterprise-class reliability features including the use of Error Correcting Code (ECC) memories ensure you won’t be compromising your system’s reliability or accuracy.

BEM applications can now be significantly accelerated with the simple addition of ClearSpeed’s enterprise-class accelerators.By accelerating standard library functions such as ZGEMM from L3 BLAS, and by providing reliable, low power accelerators that can be quickly and easily installed in your standard servers, ClearSpeed makes it easy to run bigger simulations than ever before, and dramatically reduce time-to-solution for your most demanding BEM applications.

 

Additional information:
View the CEM Market Brief