VM Performance

Has anyone benchmarked running a Tesla or a GeForce GPU on a VM either in the cloud or on their own machine? I am curious if there is a latency hit for launching kernels or pulling the data on and off. I assume most people running in public clouds don’t care about this and I couldn’t find any information on it. I’m also assuming something like Nvidia-Docker wouldn’t have any latency issues since it simply maps /dev.