checking multi-gpu environment

Thank you for your attention.

I am migrating into a new cluster environment managed by Bright/Univa Grid Engine. Each linux node has 8 K80 devices. The environment already has cuda-8.0 installed in /usr/local/cuda-8.0. If I just use that installation’s, and compile and run, for example, 4_Finance/MonteCarloMultiGPU. My session would be killed immediately.

Is there anything I can check on that node to see what caused the kill? (dmesg doesn’t seem to say anything).

Thank you very much.