Description: An initial implementation of a new iterative solver is available with this release. This new iterative solver replaces the domain decomposition iterative solver (DDM) that was available in previous releases. The new iterative solver provides improved robustness, performance, memory use, and scalability in terms of parallelization as well as model size over the DDM iterative solver.

The iterative solver should be considered only when the number of floating point operations (FLOPS) required for the direct sparse solver per iteration is prohibitive. Blocky structures with millions of degrees of freedom typically fall into this category; engineering applications that employ such models include powertrain, oil reservoir, and material microstructure simulations. While the iterative solver can provide dramatic reductions in simulation turn-around times, it should be used only in these kinds of applications due to various analysis feature coverage limitations. See “Iterative linear equation solver,” Section 6.1.5 of the Abaqus Analysis User's Manual, for more details.

The performance data presented in Table 6–1 demonstrate the strengths of the iterative solver solution for a powertrain simulation. The two problems used are identical except for mesh resolution, which results in two different model sizes in terms of degrees of freedom. The number of factorization floating point operations required for the direct sparse solver grows nonlinearly with the increase in the number of degrees of freedom. This leads to an obvious nonlinear increase in solution time for the direct sparse solver. Further increases in model size would make the use of the direct sparse solver impractical. In addition, the solution cost of the iterative solver increases almost linearly with increasing problem size. This behavior makes further growth in model size manageable. Overall, the iterative solver performs dramatically better than the direct sparse solver.

Table 6–1 Performance comparison of the iterative and direct sparse solvers for a powertrain benchmark run on 32 cores.

	Degrees of Freedom (Millions)	Direct Sparse Solver Factorization FLOPS	Direct Solver Wall Clock Time (hrs)	Iterative Solver Wall Clock Time (hrs)	Relative Speedup
Problem 1	7		0.10	0.02	5.0
Problem 2	32		3.16	0.11	28.7
Ratio	4.6	24.6	31.6	5.5	N/A

Step module:
   OtherSolver Controls