Product: Abaqus/Standard
Benefits: The new implementation of the subspace iteration method offers two significant improvements. Previously, the number of eigenmodes was restricted by the 2 GB limit imposed on the size of the subspace. The new version of the subspace iteration algorithm removes this limit. There is also significant improvement in the run time performance. In the new implementation the orthogonalization of the dynamic modes, which previously dominated the run times, has been sped up significantly by using efficient computational techniques.
Description: A new implementation of the subspace iteration algorithm delivers significant performance improvement and eliminates the restriction on the size of the subspace used. In addition, the tolerance for eigenvalue convergence has been reduced from 10–5 to 10–6 for better convergence. This may, however, increase the number of iterations.
Table 1–1 illustrates the performance improvement in the buckle step, using the new subspace iteration method. All three models were run on an 8 core, 2.27 GHz Intel Nehalem processor machine with 24 GB RAM.
Table 4–1 Performance improvement of the buckle step using the new subspace iteration algorithm.
| Model | DOF (Millions) | Number of Modes | Abaqus 6.12 | Abaqus 6.13 | Speed up | ||
|---|---|---|---|---|---|---|---|
| Number of Iterations | Wall Time (Minutes) | Number of Iterations | Wall Time (Minutes) | ||||
| 1 | 0.5 | 40 | 233 | 884 | 280 | 116 | 7.62 |
| 160 | 57 | 984 | 89 | 167 | 5.9 | ||
| 2 | 1.14 | 40 | 60 | 112 | 78 | 57.4 | 1.95 |
| 160 | 132 | 1223 | 195 | 511 | 2.4 | ||
| 3 | 0.5 | 40 | 155 | 216 | 121 | 47 | 4.6 |
| 160 | 127 | 833 | 168 | 235 | 3.54 | ||