In your machine learning workspace, you have a Compute Cluster of typeStandard_D2s_v3 with 0/2 nodes, and a Compute Instance of type Standard_D2s_v3. You want to use these computes in the most cost-effective manner, i. e. you want to avoid being charged for unwanted costs. 
Which best practice should you follow?

Answer options:

A.Stop the Compute Instance while it is unused
B.Stop the Compute Cluster while it is unused
C.For the Compute Cluster, set the number of max nodes to 0
D.Do nothing - your computes are stopped automatically when not used

Answer: A
Option A is CORRECT because the compute instances are single VMs which must be managed manually. Once you start them, they remain in “Running” state until stopped (or deleted) manually. In order to save costs, stop them when they are not in use.
Option B is incorrect because if the minimum number of nodes for the compute cluster is set to 0, the compute is automatically scales down to 0 nodes when running idle.
Option C is incorrect because the compute cluster are spinning and scaling up when they are in use, depending on the workload. If the minimum number of nodes is set to 0, the cluster scales down to 0 nodes when not in use. The maximum number of nodes has no effect on costs.
Option D is incorrect because while the compute cluster scales down automatically the minimum number of nodes, the compute instance is running generating cost even if it is idle.
Diagram - Managing compute instances 

Reference:

https://docs.microsoft.com/en-us/azure/machine-learning/concept-compute-target#azure-machine-learning-compute-managed