The rCUDA Team is developing a plug-and-play scheduler to be used with rCUDA so that multiple virtual machines can automatically (and concurrently) share the GPUs in the system without running out of GPU memory.