running cuFFTXT library on two GPUs

I am working with CUDA 6.5:

Though when I run the sample simpleCUFFT_MGPU

result = cufftXtSetGPUs (plan_input, nGPUs, whichGPUs);

I get:

CUFFT_INVALID_DEVICE An invalid GPU index was specified.

Though when I run

int nGPUs;
cudaGetDeviceCount(&nGPUs);

I get 2 CUDA Devices.

I run this code over a SLI connection.

Anybody know what is going on, and how I can resolve this?

Which GPUs are you using, exactly?

Note that CUDA (and CUFFT) have nothing to do with SLI. You are not running any CUDA or CUFFT code “over an SLI connection”.

You might want to read this section of the documentation carefully:

http://docs.nvidia.com/cuda/cufft/index.html#multiple-GPU-cufft-supported-functionality

“The GPUs must both be on the same GPU board, such as a Tesla K10 or GeForce GTX690.”

This restriction has been lifted in CUDA 7.0. The CUDA 7.0 release candidate was posted earlier today, and can be downloaded by members of the CUDA Registered Developer Program.

From the New Features section of the CUDA 7.0 release notes:

“For CUDA 7.0, cuFFT multiple-GPU execution is supported on up to four GPUs,
except for single 1D complex-to-complex transforms, which are supported on two or
four GPUs.”