AP-AMG is a high performance matrix solver using the AMG (Algebraic Multigrid) method we developed.
A matrix solver is a tool for solving linear equations for fluid dynamics, structural analysis or any other linear systems. The AMG method is a highly refined method which efficiently solves a system of linear equations by using coarse-grid based on the characteristics of coefficient matrix of a given fine-grid.
AP-AMG is a real-life industrial problem solver and is among the world's best.
|
Performances of the single version of AP-AMG is described here. The following iterative solvers compared here:
- Jacobi preconditioned CG solver (Jac-CG)
- SSOR preconditioned CG solver (SSOR-CG)
- Smoothed Aggregation-based AMG solver (SA-AMG)
- AP-AMG
The problems to be solved here are the following CFD and structural problems. "Solid" and "Shell" in the column "Model" mean solid and shell elements, respectively; "Tet4" and Tet10" mean the tetrahedral linear and quadratic elements, respectively.
| Model | No. of Nodes | DOFs /Nodes | DOFs |
Fluid dynamics | Solid_Tet4-F01 | 40,743 | 1 | 40,743 |
| Solid_Tet4_F02 | 199,450
| 1 | 199,450 |
Structural Mechanics | Solid_Tet4_S01 | 67,580 | 3 | 202740 |
| Solid_Tet10_S01 | 41,882 | 3 | 125,646 |
| Solid_Tet10_S02 | 144,544 | 3 | 433,632 |
| Solid_Tet10_S03 | 163,660 | 3 | 490,980 |
| Solid_Tet10_S04 | 450,092 | 3 | 1,350,276 |
| Solid_Tet10_S05 | 731,632 | 3 | 2,194,896 |
| Shell_S01 | 2,011 | 6 | 12,066 |
| Shell_S02 | 40,401 | 5 | 202,005 |
The results are shown below. The computer used is Xeon 2.8GHz with 4GB RAM. The maximum tolerance is set to be 10-6 for both CFD and structural problems. The mark "-" in this table shows that the memory requirement exceeded available physical memory, and could not be solved.
| Model | CPU Time [sec] |
| Jac-CG | SSOR-CG | SA-AMG | AP-AMG |
Fluid dynamics | Solid_Tet4-F01 | 10.0 | 11.5 | 4.2 | 0.7 |
| Solid_Tet4_F02 | 84 | 80 | 42 | 2.7 |
Structural Mechanics | Solid_Tet4_S01 | 118 | 72 | 39 | 13 |
| Solid_Tet10_S01 | 3,681 | 2,771 | 314 | 141 |
| Solid_Tet10_S02 | 16,178 | 9,608 | 1,446 | 350 |
| Solid_Tet10_S03 | 13,329 | 7,951 | 1,110 | 353 | | Solid_Tet10_S04 | 4,098 | 2,718 | - | 232 |
| Solid_Tet10_S05 | 17,700 | 12,605 | - | 664 |
| Shell_S01 | 21 | 13 | 5 | 3.7 |
| Shell_S02 | 402 | 244 | 67 | 31 |
|
Contact us at:
postmaster@alde.co.jp
if you have any questions.
|