Intel Pushes XenGT For GPU Access To Virtual Machines
XenGT is a full GPU virtualization solution with mediated pass-through and on the VM side runs the system's native graphics driver for the targeted hardware. When it comes to Intel hardware, Haswell's Iris Graphics is their main focus along with next-generation Broadwell processors.
XenGT is designed just not for 3D graphics acceleration within guest instances but also for media acceleration and GPGPU compute acceleration. There's use-cases for XenGT within cloud computing, data centers, rich virtual clients, multi-screen infotainment, and other areas. With other Xen GPU pass-through solutions there is no ability for both the host and guest operating systems to each access the same GPU simultaneously but they must be independently assigned at this time as there isn't a guest virtual GPU driver as in the case of VMware SVGA2 or VirtualBox Chromium. With Intel's XenGT solution, however, there is sharing support -- multiple VMs can access the same graphics processor due to its full virtualization. XenGT is pushed as offering performance, features, and sharing capabilities.
Intel made a new release of XenGT this month that has stability improvements, supports guest resolution changes, enhances support for multiple displays and hot-plugs, and there is initial support for GPU recovery. For many OpenGL workloads, the performance of XenGT is 80%+ the host's solution while for some games it was just 60%+.
Those users of Xen virtualization that want to learn more about Intel's XenGT solution that will be presented in a few hours at the Linux Foundation Collaboration Summit, the slides can be viewed right now in PDF form via this link. I'm in Napa Valley at this annual, invite-only summit so stay tuned for additional coverage later today as the more exciting talks get underway.