Wireless InSite XStream Comparisons
XStream GPU Acceleration has been extended to two of Wireless InSite’s propagation models, the X3D Ray model and Moving Window FDTD (MWFDTD) model. Calculations can now be run on NVIDIA’s CUDA-enabled GPUs, dramatically improving speeds.
Moving Window FDTD performance enhancement and speed comparisons…
X3D Ray Model performance enhancement and speed comparisons…
Moving Window FDTD Performance Enhancement and Speed Comparisons
Moving Window FDTD (MWFDTD) is a full-wave model for propagation of radio waves over irregular terrain. Wireless InSite Release 2.6 offers a new implementation of the model that dramatically reduces the calculation time. Changes to MWFDTD include:
- A new C++ implementation of MWFDTD
- XStream GPU Acceleration has been extended to MWFDTD, allowing calculation to be run on NVIDIA’s CUDA-enabled GPUs.
- Variable Grid Option – The FDTD computational grid can now be constructed with variable cell spacing, providing accurate results with fewer cells.
Compared to previous releases, MWFDTD projects using Release 2.6 on a CPU show speed improvements of 5.5 to 17.2 times. When run on a GPU, MWFDTD runs 38 to 113 times faster than previous releases.
The new MWFDTD can now provide high fidelity to a new class of scenarios that were previously prohibitive.
Figure 1: Comparison of MWFDTD calculation times
Figure 2: MWFDTD speed-up relative to Release 2.5
Table 1: MWFDTD Release 2.5 and 2.6 calculation times.
Timing comparison using an Intel Xeon X5660, 2.80 GHz CPU, and an NVIDIA C2050 GPU
X3D Ray Model Performance Enhancement and Speed Comparisons
The X3D Ray Model is Remcom’s first ray-based model to use XStream GPU acceleration and multi-threading to reduce calculation time. Depending on the type of project, X3D run times are 2.5 to 15.1 times faster than the Full 3D model, as demonstrated in the comparison figures below.
For these comparisons, all projects included 10 reflections, 0 transmissions, and 2 diffractions. For the timing comparison, an Intel Xeon X5660, 2.80 GHz CPU, and an NVIDIA C2050 GPU were used.
Figure 3: Calculation times for X3D and Full 3D
Table 2: Calculation time comparison between Full 3D and X3D ray models
