Intra-node Memory Safe GPU Co-Scheduling