Difference between revisions of "EM.Tempo Acceleration"
From Emagtech Wiki
Line 1: | Line 1: | ||
− | [[Image:FDTD58.png|thumb|300px|[[FDTD Module]]'s Engine Settings dialog]] | + | [[Image:FDTD58.png|thumb|300px|right|[[FDTD Module]]'s Engine Settings dialog]] |
{{version|14.8+}}[[EM.Tempo]] presents three different options for acceleration. Some details about each of the options are discussed below: | {{version|14.8+}}[[EM.Tempo]] presents three different options for acceleration. Some details about each of the options are discussed below: | ||
Revision as of 16:01, 29 August 2014
This content is relevant for EM.Cube: | 14.8+ |
- Single-Precision Solver: EM.Tempo's single-precision solver is a multi-threaded FDTD engine which also takes advantage of modern SIMD facilities (e.g. Intel AVX) found on most processors. It trades a slight reduction in overall accuracy for a relatively large boost in performance.
Accuracy-critical computations, such as Farfield calculation, are still performed in double-precision. |
If EM.Cube determines your system does not have the appropriate SIMD facilities to effectively run the single-precision solver, the double-precision solver will be run automatically. |
- Double-Precision Solver: EM.Tempo's double-precision engine is also multi-threaded, and is slightly more accurate in some cases. It does not take advantage of any available SIMD facilities, which makes it a very portable solver.
- GPU Solver: EM.Tempo's GPU solver is a CUDA-based engine which is capable of extreme performance (speedup of up to 40x).
Before attempting to use EM.Tempo's GPU engine, please make sure all drivers are up to date. |
It is critical to have the right environment to efficiently use the GPU solver. For immediate help, please contact collaborate@emagtech.com |
Thread Factor
The thread factor determines how many threads are effectively used in the computation. It is only relevant for CPU-based solvers. While a thread factor of two or three usually yields the best performance, it is suggested that the user experiment to find the best thread factor for his/her system.