I'm about to launch 10 clients (30 video cards) online to do some GPU mining at a colocation facility on a full rack. I would like to simplify the management of the systems by PXE booting from a PXE server hosting the PXE image. I would be using a ramdisk so that I can perform updates on the PXE image and have it applied to all the clients at once. Would this be the optimal solution?
The clients would be running the bare minimum in Ubuntu to get GPU mining up and running with a slight overhead. How much memory do you think I need?