I am also interested in the answer to this question. I would also like to know if the graphics cores can also be a virtualized resource, as the A15s should be. I posed this question and obtained the following response:
"the Jetson TK1 Kit should support virtual machines. However, in regards to the setup and the functionality please post your query in the developer zone Forums where you can get the accurate details :
NVIDIA Customer Care"
My followup questions for NVidia engineers are:
What flags should one look for to determine if virtualization extensions are available?
Does the board support package/bootloader support startup in ‘Hyp’ mode?
Must the serial console be connected and u-boot options be selected to boot into ‘Hyp’ mode?
Does u-boot need to be recompiled and reflashed to enable booting in ‘Hyp’ mode? (Note that the “NVIDIA TEGRA LINUX DRIVER PACKAGE” PDF file (at https://developer.nvidia.com/linux-tegra-rel-19) has instructions for building the BSP and OS, including u-boot.)
There is further reason to believe that Tegra K1 would support virtualization of the A15 cores, based on the following:
http://on-demand.gputechconf.com/gtc/2013/presentations/S3577-Tegra-Virtualization-in-Automotive-Applications.pdf discusses an automative application without hardware assisted virtualization of Linux instances running on Tegra 3 and makes mention of hardware-assisted virtualization in A15s. So, to support automotive applications and reduce processor count while serving all systems and providing ‘infotainment’, it seems reasonable that the A15s of the Jetson TK1 would each support hardware assisted virtualization to provide isolation and reduced processor count.
Also, note the state of the art for virtualization options with CUDA gfx cores (from http://www.acceleware.com/blog/state-gpu-virtualization-cuda-applications-2014):
Pass-Through is the most common mode for GPU virtualization today. It is supported by VMware, Oracle, Red Hat and Citrix. Citrix is by far the leader in GPU virtualization technology with the ability to take advantage of the NVIDIA Grid platform. For now, Microsoft is lagging with no support for CUDA in Hyper-V. GPU Virtualization is advancing rapidly with major improvements expected. In 2015, VMware will support GPU partitioning. Red Hat’s roadmap includes GPU oversubscription with no timelines. Looks like next year’s survey will have a lot more to cover!"
Seems like, at best, GPU virtualization is for GRID applications in a datacenter…no surprises. However, it might be possible to test out pass-through of the GPU to a VirtualBox VM, as Virtualbox runs under Linux (A flavor of Ubuntu for the TK1). Xenserver would replace Ubuntu on the TK1 and render it unable to then perform its inherent magic with Linux 4 Tegra; so that is not an option.