I have Nvidia 8800 GTS on my 2.8GHz system.
I tried the CuFFT library with this short code.
It works fine for all the size smaller then 4096, but fails otherwise.
cufft: ERROR: D:/Bld/rel/gpgpu/toolkit/r2.1/cufft/src/config.cu, line 331
cufft: ERROR: CUFFT_ALLOC_FAILED
cufftSafeCall() CUFFT error in file <simpleCUFFT.cu>, line 193
void runTest(int argc, char** argv)
{
int NX = 4096;
int NY = 4096;
if( cutCheckCmdLineFlag(argc, (const char**)argv, "device") )
cutilDeviceInit(argc, argv);
else
cudaSetDevice( cutGetMaxGflopsDeviceId() );
Complex* h_idata = (Complex*)malloc(sizeof(Complex) * NX * NY);
cufftHandle plan;
cufftComplex *idata, *odata;
cutilSafeCall( cudaMalloc((void**)&idata, sizeof(cufftComplex)*NX*NY));
cutilSafeCall(cudaMalloc((void**)&odata, sizeof(cufftComplex)*NX*NY));
// Copy host memory to device
cutilSafeCall(cudaMemcpy(idata, h_idata,sizeof(Complex) * NX * NY, cudaMemcpyHostToDevice));
/* Create a 2D FFT plan. */
cufftSafeCall(cufftPlan2d(&plan, NX, NY, CUFFT_C2C));
for(int i=0;i<10;i++)
/* Use the CUFFT plan to transform the signal out of place. */
cufftSafeCall(cufftExecC2C(plan, idata, odata , CUFFT_FORWARD));
cufftSafeCall(cufftDestroy(plan));
cutilSafeCall(cudaFree(idata));
cutilSafeCall(cudaFree(odata));
}