© 2012 QWED Company. All rights reserved.    2012  |  Home  |  Events  |  Products  |  Applications  |  EU Projects  |  About  |  Support  |  Log In
 
QuickWave - QW-AccelSim comparision
Computer configuration: 2 x Dual Core AMD Opteron 2GHz, 4GB of RAM, Windows XP Professional x64. Results are presented for 1000 iterations.
Beefburger
This is a standard beefburg.pro example from Qw_examp/QW_3D/Standard/ Beefburg directory. Its basic meshing is 1.7 mm x 1.7 mm x 1.7 mm, which with several refined / coarser mesh regions leads to 1.3 Mcells. For speed benchmarking purposes, mesh snapping planes have been subsequently removed and discretisation has been uniformly refined. The finest case of 0.937 mm x 0.937 mm x 0.937 mm leads to 24 Mcells.
The figures below show various measures of speed for a successively refined mesh:
time for 1000 iterations
number of FDTD iterations per second
number of FDTD cell updates per second
speed-up of QW-AccelSim with respect to QW-Simulator on CPU
Dielectric resonator
This is a standard dielres1.pro example from Qw_examp/QW_3D/Standard/Dielres directory. Standard mesh is 2 mm x 2 mm x 2 mm (0.03 Mcells). For speed benchmarking purposes FD-Monitor postprocessing has been removed (as it is not currently supported by QW-AccelSim) and mesh has been uniformly refined. The finest meshing case is 0.194 mm x 0.194 mm x 0.194 mm (25 Mcells).
For commensurate problem sizes, the speed up of dielres1.pro is better than beefburg.pro. There are several factors:
beefburg.pro contains a metal region above the oven ceiling, next to the waveguide feed, where the fields are not updated anyway in QW-Simulator (and hence effective number of cells is lower)
dielres.pro requires field exchange between CPU and Acceleware card only at a single point (for punctual excitation and FD-Probing), while beefburg.pro requires field exchange in the waveguide cross-section (template excitation and S-parameters postprocessing)
SOFTWARE FOR ELECTROMAGNETIC DESIGN
enjoy the simulation