Train the model for a small amount of data Leverage the compute storage to see how fast the model will train on an architecture with GPUS Read all the data for the past year Install gpu driver on the compute