parallel workers cannot see GPUs

I'm currently working on code that should run on a number of different HPC machines with multiple GPUs. The machine I'm testing and developing my code hast just one GPU. The idea is to create a worker for each GPU found on the machine and then execute via a parfor loop some tasks on the GPUs. If I query gpuDeviceCount I get a 1 back but if I execute

    parpool(1) % same is true for any number of workers
    spmd 
        gpuDeviceCount
    end

I always just get 0s back. Why are my workers unable to see the GPU?

[Walter Roberson]

Potentially relevant:

http://blogs.mathworks.com/loren/2013/06/24/running-monte-carlo-simulations-on-multiple-gpus/

http://www.mathworks.com/matlabcentral/answers/24019-gpu-computing-in-spmd

[Subhra Saha]

The below link might be helpful:

http://www.mathworks.com/matlabcentral/answers/98712-why-am-i-not-able-to-access-a-gpu-card-on-a-windows-vista-windows-7-machine-managed-under-job-manage