When a descriptor or algorithm setup for a particular FFT size and type already exists, the computation of the FFT can be much (3-5x) faster. However, if the number of different FFT sizes and consequently descriptors is too big, the system can run out of memory. Notice also that FFT's of lengths other than power of two, can be 3-5x slower. As a guideline, allocating 100 descriptors for different (1D) FFT sizes in range from 39600 to 40000 will require around 300MBytes of memory. The bigger the FFT's that you compute, the smaller must be the descriptor cache size and fewer different FFT sizes should be computed to keep the performance level high. 

Note: This property should not be changed while there are FFT's being executed!

TMtxVecController Properties