It does not have to be. We have optimizations for all kinds of workloads - https://CentML.ai