On Windows systems the manner in which many I/O tasks are performed for ABAQUS/Standard has been tuned to improve I/O performance. Smaller jobs (with fewer than 100,000 degrees of freedom) should use more than 85% of the CPU's processing power during an analysis in Version 6.6; similar analyses in Version 6.5 could use as little as 50% of the processing power. This improvement reduces processing time for such analyses.

When you run a job with standard_memory_policy=MAXIMUM, ABAQUS does not write the factorized stiffness matrix to disk, provided that the memory value specified by standard_memory is high enough to allow the factorized stiffness matrix to be held in memory. This enhancement provides the greatest benefit when jobs are run in parallel on systems that do not have high-performance file systems or on machines where I/O performance is generally a bottleneck.

ABAQUS/Standard also provides a new direct solver algorithm that significantly reduces the computational cost of factorization in analyses of models with one or more large distributed couplings. This feature is available for all ABAQUS/Standard analyses except eigenvalue extraction procedures (such as natural frequency extraction or eigenvalue buckling) and substructure generation analysis steps.

ABAQUS/Standard now uses a different default method to estimate memory requirements for analyses that use the penalty method without Lagrange multipliers to enforce contact. For problems that use penalty contact without Lagrange multipliers, ABAQUS/Standard now calculates memory estimates based on the initially active contact constraints. For non-penalty analyses and for similar problems in previous versions, ABAQUS/Standard estimates memory requirements based on a conservative estimate of potential contact interactions over the course of the analysis. This new approach provides more accurate estimates of memory requirements.

In addition, the memory required by the equation solver is often reduced significantly by avoiding Lagrange multipliers. The memory estimates for non-penalty analyses can be quite conservative; that is, they tend to overestimate actual memory needs. The memory estimates for analyses that use penalty contact are not always conservative for the entire analysis; for example, the memory needs may exceed the initial memory estimates if most contact constraints are initially “open” (inactive) but “close” (become active) during an analysis. If you set standard_memory equal to a memory estimate that turns out to be non-conservative, you may need to rerun the analysis with the standard_memory setting increased.