Maximum Number of FFTs in Parallel


According to the CUDA C Programming Guide Version 4.0 (5/6/2011):

Now, if I use cufftPlanMany() to perform multiple FFTs in parallel, does the above limitation mean that I can only perform up to 16 FFTs in parallel? The CUFFT library documentation (PG-05327-040_V01, February 2011) does not mention the maximum number of FFTs that can be done in parallel. (Or is it unrelated to the quoted statment, and dependent only on the length of the transform and the specs of the GPU?)


It is unrelated to the number of concurrent kernels.

Thank you!