why am i doing all of this? i have a bursty GPU-based application, so it's best to have a scalable + on premises fixed compute k8s cluster