What are capacities?
Every operation needs compute power. For each workload in Fabric, the optimal type of compute may look slightly different. Where SQL runs best on CPUs, Spark will be better on GPUs. Some workloads, such as Power BI, require large amounts of memory, while others such as Data Factory may just need small amounts of CPU for orchestrating activities. Fabric capacities remove the complexity of choosing the right type and size compute for each operation that is run by providing a single, blended compute metric called a capacity unit that is delivered through Fabric capacities, often referred to as a Fabric SKU or simply an F SKU.
Therefore, Fabric capacity is simply a pool of available resources that is shared by all the workspaces assigned to a capacity and is shared by all the workloads running in those workspaces. Based on the type of operation running, the appropriate blend of compute and memory will be allocated and torn down after the operation is complete because...
