In practice this will probably work and maybe useful in a test lab. However this should not be used in production as it is currently an uncertified solution. Both OEMs (server vendors) and virtualization stacks (Citrix/VMware) do not currently test or certify servers and GPUs for this combination so it is not a configuration NVIDIA can support or endorse for enterprise customer who want their stack supported.
Whilst customers are free to try this at their own risk, NVIDIA is unable to provide support for problems arising within systems that are not tested nor certified by OEMs and virtualization stacks.
Customers must refer to their virtualization vendor's HCL (Hardware Compatibility List) for GPU technologies such as GPU pass-through, vGPU, vSGA and vDGA technologies.
For physical deployments customers should consult NVIDIAs own GRID certified server list, which currently _does not_ test or certify any servers with mixed card combinations such as:
· K1 cards mixed with K2
· M60 cards mixed with K2
· VMware vSphere HCL/HYPERLINK "https://www.vmware.com/resources/compatibility/search.php?deviceCategory=vdga"ESXiHYPERLINK "https://www.vmware.com/resources/compatibility/search.php?deviceCategory=vdga" HCL
· VMware GRID vGPU FAQs (section on certified servers): http://www.vmware.com/files/pdf/products/horizon/vmware-nvidia-grid-vgpu-FAQ.pdf
NVIDIA GRID bare-metal, vGPU, GPU-passthrough, vDGA and vSGA