On Sun, Apr 30, 2017 at 8:51 PM, Nick S <nick.kv...@gmail.com> wrote:
> I know there is a project for Intel graphics. Anybody knows whether we > have any support for the new GPU virtualization tech from either NVidia of > AMD? I know I can pass the entire card to the guest, my question is whether > some GPU virtualization is possible on one of these platforms where I can > share the card between multiple guests. (Owner of a 4 GPU box here running > out of PCI-e slots) > For NVIDIA, yes, Linux kernel v4.10 includes the VFIO Mediated Device framework which was a collaborative development with NVIDIA, Red Hat, and Intel to add vGPU support for both vendors. You'll need at least QEMU 2.7 for certain features required by the vendor drivers. libvirt 3.2.0 also introduced basic support for management of VMs with pre-defined mdev devices. You can look at NVIDIA's documentation for their current Xen vGPU offering to get an idea of which hardware products will be supported with this (not GeForce). I cannot speak to software availability from NVIDIA. For AMD, theoretically they've taken an SR-IOV approach, so each vGPU would be a typical Virtual Function. Ideally this would require no special code for KVM/VFIO. SR-IOV does however require support in the Physical Function driver to enable the VFs and I don't have any details on whether this code is upstream yet. I have no experience, direct or indirect, with AMD's offering here. Alex
_______________________________________________ vfio-users mailing list vfio-users@redhat.com https://www.redhat.com/mailman/listinfo/vfio-users