By using taints and tolerations workloads can be scheduled onto specific node pools, allowing just the right amount of the required special resources to be used when required.

without this capability our entire cluster would need to run with a SSD or GPU enabled SKU, forcing our COGS to increase.