Scaling up means making one warehouse bigger. Scaling out means adding more clusters of the same size, called a multi-cluster warehouse. They fix different problems.
Scale up: for slow queries
A bigger size (for example Medium to Large) gives each query more CPU, memory and local disk. Each size step doubles both the machines and the credits per hour. Use it when a single query is slow because it is heavy: big joins, large sorts, or "bytes spilled to local or remote storage" in the profile. A larger warehouse often runs a query about twice as fast for twice the rate, so the total cost can stay about the same while the user waits less. That is true only if the query scales well.
Scale out: for many users
Multi-cluster warehouses (an Enterprise edition feature) start extra clusters when queries begin to queue, and shut them down when the load drops. You set a minimum and maximum number of clusters. Use it for BI dashboards at 9 a.m. when 80 people open reports at the same time. Each query still runs at the same speed, but they stop waiting in line.
Slow single query? -> size up Many queries waiting? -> multi-cluster (scale out)
Making a warehouse bigger does not help with queueing, and adding clusters does not help one slow query. Mixing these up is a common interview mistake.
Controlling the cost
- Auto-suspend: stop the warehouse after a short idle time (a minute or two is common for ad-hoc work).
- Auto-resume: start it again on the next query.
- Billing is per second, with a 60-second minimum each time the warehouse starts. So suspending after 5 seconds of work and resuming often can still cost a full minute each time.
Also remember that suspending clears the local disk cache, so the next queries read from remote storage and can be slower at first.