CUFFT 1D Memory Limitations


I am running into some unexpected memory limitations
with 1D FFTs using CUFFT in batched mode. Using a 512MB
card the batch size limit seems to be about 180MB worth of
data for vectors of size 16384. Up to that point everything works
fine and the outputs within tolerance.

Above this point, ExecC2C fails. Why is it failing? Will a GPU
with 2x as much GPU RAM take 360MB? Any ideas?