I am currently helping to organise a training day for programming using the Xeon Phi. We have a Linux shared memory machine that has 31 MIC/Xeon pairs and we'll create temporary accounts on the machine for each of the trainees. So we want to divide up the number of MIC cards evenly and allow each user to be able to offload to just 1 MIC card.
Is there a simple way to do this through the environment? There is this OFFLOAD_DEVICES variable I found in the MKL documentation, but is that only for automatic offloading of MKL routines?