News
GPU Acceleration Of Electromagnetic Solutions In FEKO
May 24, 2010
Complex solution algorithms as available in FEKO (e.g. MoM, PO, UTD, FEM, or MLFMM) were historically not well suited to acceleration with GPUs (graphical processing units). These worked best for FDTD or similar, where the same operation is repeated over and over again. However, recent advances in software technology have made it possible to accelerate run-time intensive solution phases in FEKO (to be available in the next release FEKO Suite 6.0).
Recently high-performance computing has seen a boost through the usage of parallel processing capabilities of multi-core GPUs (graphics processing units). Even fairly standard graphics cards found in typical engineering desktop PCs have computing capabilities exceeding those of high-end CPUs.
EM Software & Systems – S.A. (Pty) Ltd (EMSS) is proud to announce that in the next release of its electromagnetic solver FEKO Suite 6.0 GPU and mixed CPU/GPU processing will be integrated for run-time critical phases of the solution process of some of the solvers integrated in FEKO.
For moderate to large size Method of Moments (MoM) problems, the most run-time critical solution phase is the solution of the system of linear equations. Even though FEKO employs highly optimised libraries for this phase for the various CPUs, the GPU capabilities can beat this performance by more than one order of magnitude.
As example, we have studied the radar scattering from a metallic object over a frequency band with frequency dependent meshing (i.e. using between 2 700 and 10 000 unknowns) and compare the time for the matrix solution phase (i.e. excluding times for matrix setup, near- and far-field etc.). Fig. 1 shows that for an NVidia GTX295 graphics card (using one GPU only) the time is 13.9 sec, while on the three CPUs tested for comparison the times vary between 122 and 397 seconds (i.e. GPU is a factor of 9 to 29 faster than these CPUs).
Often the performance is measured in GFLOPS (billion floating point operations per second), and for the same radar scattering problem the achieved FEKO performance for this solution phase of solving the MoM system of linear equations is depicted in Fig. 2. Also from this graph the superiority of GPU processing for such types of calculations is obvious.
Note that the results presented here might further improve (optimisations) by the time of the release of FEKO Suite 6.0. It is also expected that the release of new GPU hardware will greatly impact performance (for instance for double precision algebra a significant boost is expected by NVidia's Fermi architecture). It is also planned to have other solution phases of FEKO for other techniques accelerated by GPUs.
