Providing vGPUs to VM on QEMU/KVM

Is there a stable solution to providing vGPUs to VM on QEMU/KVM?
Customers which are openstack/kvm based are willing to use vGPU with NVIDA tesla M10, M60 in there VMs. I know XenServer and Vmware horizon is the good idea, but what about on openstack/KVM?
I have found a testing solution "", but seemly it is not ready for production use.