I have some custom analysis software written with CUDA. I’d like to host this software in a virtual machine in VMWare. VMWare allows for “pcipassthrough” to give direct access to a PCI card (such as a GPU) and I’ve read that the Quadro 4000 GPU will work for a CUDA application through VMWare with “pcipassthrough”. My question: Has anyone successfully used a Quadro 4000 or any any other NVIDA GPU card with a CUDA application hosted on a VMWare VM with “pcipassthrough”?
Make sure your CPU supports VT-d, if it’s already running VMWare in some sort of enterprise environment it probably does. VT-d is a requirement for the passthrough feature.
I was able to do this in ESXi 5.5 with a slightly different card:
I can only assume it would work with VMWare, but have not tried that route.