max 2d cufft dimension

Hello,

I’m a little bit confused with a sentence of the cufft documentation:

“2D and 3D transform sizes in the range [2, 16384] in any
dimension.”

So in my testing application I’m trying to do a 2D R2C forward , and right after that a 2D C2R inverse fourier transformation, to receive the source data. As I try bigger and bigger testing data I assumed that I would be able to transform 16k x 16k float data, according to the docu.

1kx1k data works , but the next step 2kx2k i get the following error:

cufft: ERROR: /root/cuda-stuff/sw/gpgpu_rel1.1/cufft/src/plan.cu, line 35
cufft: ERROR: CUFFT_INTERNAL_ERROR

So is it just that I’m out of Memory ( ‘2k x 2k x float’ is 16 M + double for the cufftComplex . I think this should be no problem ), or any other Problem.? (suggestions ?) Both transforms are out-of-place transforms that consider the non-redundant data fact (only for the memory issues). ( N x N -> N x N/2 for the forward and vice versa for the inverse)

My tessting environment is linux and a 8800 GTS with 640 MB.

I hope maybee someone can just help me with some miss understanding I have.