Burst-core error vs timestep
All published timestep points · lower error is better
TDSE OnTDSE Off
Benchmarks
Measured, not assumed.
Runtime performance across hardware platforms. Numerical accuracy against declared references.
01 / Runtime performance
Small steps. Measured in microseconds.
25 platform groups · 28 execution configurations · 1299 matrix cases.
Windows 11 · AMD Ryzen 7 9800X3D (8 CPU cores)
5.68µs
per prepared step Measured median · lower is faster
32 ports · 1,024 history samples · FP32
Compact FP32np 8 · nh 512 · FP32
2.21 µs
Mid-size FP32np 32 · nh 1024 · FP32
5.68 µs
Mid-size FP64np 32 · nh 1024 · FP64
11.86 µs
Large FP64np 128 · nh 2048 · FP64
3,841.08 µs
np: port count · nh: history length · FP32 / FP64: numeric precision
CPU: host elapsed time. GPU: device-resident vectors, including completion wait. FPGA: host-observed step with vector transfers. Different timing boundaries are not a cross-family ranking.
What these measurements cover
The report covers 16 CPU, 8 GPU and 1 FPGA platform groups. The selector follows PDF order; multi-GPU execution configurations remain in the complete report.
These are measured prepared-step medians, not complete simulator runtimes or real-time deadline guarantees. Missing representative points are shown explicitly.
Read the complete runtime report02 / Numerical accuracy
Numerical accuracy against reference solutions.
Explore frequency and timestep sweeps, two-port excitation, a short-fault case and cross-system evidence, each against its stated reference.
Burst-core error at fine timesteps
Published points from 5 to 30 µs · lower error is better
TDSE OnTDSE Off
Exact published values at the crossover
| Step (µs) | TDSE On | TDSE Off |
|---|---|---|
| 20 | 0.07758922454029925 | 0.07346179028020827 |
| 22 | 0.07347479817253319 | 0.08896716638634201 |
Complete published timestep range. Switch to fine steps to inspect the crossover.
IEEE118 / Bus 12 / One port
The timestep changes the comparison.
TDSE Off is closer to the reference at fine timesteps. The curves cross between 20 and 22 µs; beyond that, TDSE On has lower mean burst-core error in this study.
2000 Hz Gaussian burst · 22 timestep settings · 50 trials per setting · 1 µs gold reference. Mean relative error within the burst core, not a percentage.
Inspect the setup and complete studyEach case keeps its own reference, scoring window and limitations. Fine-step regimes where TDSE Off is more accurate remain visible; the results are not combined into a single accuracy score.
Read all public accuracy resultsBring your own acceptance criteria.
Accuracy, preparation cost, host overhead and deployment constraints belong in the same evaluation.
Discuss an evaluation