I have been using Compute engine with an instance to which I attached Tesla K80 GPU. I notice,occasionally sharp performance drops of the graphics application I am running there. The app is always the same.Its workload doesn't change. So I suspect it has something to do with the current load on the hardware from other places. How does it work with Google Cloud regarding Compute engine instances? If the hardware is shared by multiple users, then is there a way to reserve it (even with extra costs) for my usage only?
You could track your CPU steal time by watching at %st on TOP command.