cufft_exec_failed on the second gpu cufft_exec_failed multi gpu

My machine has 2 cards, first is Tesla C2050, second is FX 1800.

When I call cufft , no exception on C2050, but I get a cufft_exec_failed error on FX 1800.

Furthermore, I found it’s fine for some sizes,I run a test on FX1800, when size less than 320 it works fine,

and when size greater than 320 may get cufft_exec_failed error for certain sizes

So who can tell me why??? External Image

cudaError_t err = cudaSetDevice(deviceId);  //here deviceId = 1   FX 1800

cufftComplex *d_x ;

cudaMalloc( (void *)&d_x,sizeof(cufftComplex)*len );

// cudaMemcpy  host to device

cufftHandle plan;

cufftResult re = cufftPlan1d(&plan,len,CUFFT_C2C,1);

re = cufftExecC2C(plan,d_x,d_x,CUFFT_INVERSE);

// cudaMemcpy device to host

cufftDestroy(plan);

cudaFree(d_x)

printf("fft error = %d\n",re);