Benchmark1: CAD model of car, driver, cell phone with overlaid electric fields of the simulation
Benchmark1: Electric field in phone and hand (detail)
Benchmark2: Human male head, deep brain stimulation implant and box
Benchmark3: Full CAD PCB board
PERFORMANCE
SEMCAD X is setting new standards in CEM by offering some of the fastest FDTD solvers and enhancements on the market. Offering awesome speedup, excellent RAM memory efficiency and useful features like auto-termination, solving large problems has never been so easy.
Furthermore, SEMCAD X was the first FDTD toolkit in the market offering hardware acceleration. Simulation speeds of 700-3000 MCells/s puts SEMCAD X in a league of its own as far as solver performance is concerned.
Please do contact us to request performance information related to your specific applications. The following 3 examples are used as performance benchmarks to highlight the advanced performance of SEMCAD X:
| SEMCAD X Simulation | Bench 1 | Bench 2 | Bench 3 |
| Model | car, driver, phone | head with implant | PCB |
| frequency (MHz) | 1800 | 64 | 500-2000 |
| No. time step | 5896 | 406435 | 404530 |
| computational domain (million cells) |
296.6 | 355.8 | 286.2 |
| excitation |
antenna | resonator | coax line |
| ABC | UPML 8 lyrs | UPML 11 lyrs | UPML 9 lyrs |
| Solver Performance | Bench 1 | Bench 2 | Bench 3 |
| GPU Solver Speed (Mcell/s)* | 2994 | 2685 | 2524 |
| GPU Solver Time (hh:mm) | 1:04 | 28:54 | 71:15 |
| CPU Solver Speed (Mcell/s)** | 27 | 19.8 | 15.6 |
| CPU Solver Time (hh:mm) | 52:21 | 2043:00 | 280:00 |
** HPxw9400 workstations with Dual Core AMD Opteron 2216 at 2.4 GHz and 16 GB of RAM
The following table shows the performance of SEMCAD X and Acceleware's FDTD solver using NVIDIA Series 20 (Fermi) hardware.
| NVIDIA GPU | Number of GPU’s | Maximum Speed Achieved in Mcells/s | Optimum Simulation Size in Mcells |
| Tesla C2070 | 1 | 780 | up to 150 |
| 2x Tesla C2070 | 2 | 1650 | up to 280 |
| Tesla S2070 | 4 | 2800 | up to 560 |
| 2 ... n Tesla |
n * 4 | > 3000 | > 1000 |
NVIDIA's Fermi architecture adds improved double precision throughput, ECC support, and improved overall performance. In addition, two or more of these cards can be connected in parallel to allow higher performance and larger domain sizes. The Series 20 shows remarkable improvement over the previous Series 10 (Tesla), but it positively eclipses the solver speed of CPU-only solver. Solving large-scale, high-resolution problems in CPU-based software becomes too burdensome to be practical, as shown in the plot below.
Comparison of Simulation Size vs. Solver Speed in Fermi hardware
SEMCAD X's numerical solver allows multiple computers as a Beowulf-style cluster to cooperatively run a single FDTD simulation using the Acceleware Cluster library. A single simulation space is partitioned across multiple computer nodes using MPI processes and time stepped synchronously. This cluster CPU solution is now available Linux 64 bit architectures.
The following benchmark was run in the Ben Arabi cluster at FPC Murcia, Spain. The Arabi cluster consists of 102 nodes, for a total of 816 cores of Intel Xeon Quad-Core E5450@3GHz and 1072 GB of distributed memory. The benchmark used was the generic phone model gridded at 195 Mcells. This example shows how solver speed sclaes with the number of cores executing the simulation.
