K10 error behind CPU1 in multi-cpu system

cual cpu fails with a K10 error if the K10 is installed behind CPU1 but the same software and K10 work when the K10 is installed behind CPU0. Here’s customer feedback

Unable to determine the device handle for GPU 0000:89:00.0: The NVIDIA kernel module detected an issue with GPU interrupts. Consult the “Common Problems” Chapter of the NVIDIA Driver README for details and steps that can be taken to resolve this issue.

The customer then ran matrixMul and got the following fatal error:
cudaGetDevice returned error code 10, line(396)
cudaGetDeviceProperties returned error code 10, line(409)
MatrixA(160,160), MatrixB(320,160)
cudaMalloc d_A returned error code 10, line(164)

Then customer ran transpose and got the following fatal error:
CUDA error at …/…/common/inc/helper_cuda.h:905 code=10(cudaErrorInvalidDevice) “cudaGetDeviceCount(&device_count)”

Just for “fun”, the customer tried putting an NVidia Quadro K4000 behind the second cpu to see if it would work as a normal video card, and it did.