Cloud computing technologies have matured and expanded dramatically in recent years. Along with this expansion, DoD organizations are being driven to take advantage of this technology and host as many computing systems as possible from Core Data Centers. This makes sense in many regards, except when it comes to graphically intensive applications. Virtual Machines (VMs) perform very poorly when having to virtualize the more advanced abilities of hardware Graphical Processor Units (GPUs) into software. Other technologies have been introduced to attempt to counter this deficiency, namely, graphics cards that have GPUs that can be divided into separate GPU segments, called “vGPUsâ€� that are then directly assigned to VMs. These technologies offer the promise of being able to host even graphically intensive applications in a cloud environment. However, how well do these combined technologies perform when hosting demanding simulations? Can existing simulation systems be migrated to such a cloud environment? Are the technologies sufficiently mature enough to meet current performance demands? How well do they perform under realistic network conditions? To answer these questions, the PEO STRI Project Manager for Integrated Training Environment (PM ITE) teamed with the MITRE Corporation to conduct a series of experiments to evaluate the feasibility of hosting graphically intensive simulations within a hypervisor/cloud environment using advanced vGPU technologies. This paper reports on how the experiments were configured and executed, and presents a variety of performance observations including draw rates, CPU and GPU utilization, network bandwidths, etc. The paper also discusses what trade-offs were evaluated in order to host the highest number of VMs from a single server with the most acceptable performance, as well as the impact of operating within realistic network conditions. Finally, implications for potential future use of the technology are discussed.