Performance Optimization CPU ==== Linear solver parallelism ---- The modal calculation uses a linear solver. If we choose the linear solver MUMPS with the keyword factor SOLVEUR, operand METHODE =' MUMPS ', we can activate its parallel operation. To do this, you must use a parallel version of *Code_Aster*, and enter the number of processors to use. Empirically, the use of one processor per slice of 105 to 106 degrees of freedom in the model provides good performance. At the same time, we also recommend setting RENUM =' QAMD '. Calculation of modes by sub-bands ---- If the problem to be treated is a GEP and you are looking for a lot of modes on a given band or if the number of degrees of freedom of the model is high (> 105), it is advisable to split the global search band into several sub-bands, from 40 to 80 modes, as balanced as possible. The calculation will thus be more robust and faster, even when sequentially calling a CALC_MODES for each sub-band. For better ergonomics, calculations on each sub-band can be performed automatically and transparently for the user, by giving the keyword FREQ a list of n>2 values. This also makes it possible to parallelize the processing of the various sub-bands, generating considerable gains in performance CPU (a factor that can reach 10 to 20 on the calculation time, and several tens of percentages on the memory peak). For more details, refer to the generic documentation [:external:ref:`U2.08.06 `] on parallelism, and [:external:ref:`U2.06.01 `] on the calculation of natural modes.