Matching a workload to the right compute service and its cost model

Databricks offers several compute types—All-Purpose, Job, SQL Warehouses (Serverless, Pro, Classic), and Serverless compute for notebooks/jobs—each optimized for different workloads and billed differently in DBUs. Choosing the right compute impacts cost, startup latency, and concurrency, so exam scenarios test matching workload characteristics (interactive dev, scheduled ETL, BI/SQL queries) to the appropriate compute type.

1 · Learn the must-know

• 1All-Purpose compute is for interactive development in notebooks (collaborative, higher DBU cost) while Job compute is auto-created/terminated per job run and is cheaper per DBU since it lacks interactive overhead.
• 2SQL Warehouses (Serverless, Pro, Classic) are optimized for SQL analytics/BI tool connections; Serverless SQL Warehouses start almost instantly (no cluster spin-up wait) and auto-scale, trading some cost for speed and simplicity.
• 3Serverless compute (for notebooks, jobs, and SQL) removes the need to configure/manage clusters, reduces startup latency to seconds, and shifts billing purely to DBUs consumed rather than cloud VM uptime.
• 4Classic (non-serverless) compute requires you to configure instance types, autoscaling, and cluster policies, and you pay for underlying cloud VM time in addition to DBUs, so idle/misconfigured clusters can waste cost.
• 5Job compute should be used for scheduled/production pipelines to minimize cost since it's cheaper than All-Purpose and clusters are ephemeral, terminating automatically after the job completes.
• 6Cluster policies and autoscaling/auto-termination settings are key cost-control levers regardless of compute type, and exam questions often test recognizing when a workload is using an unnecessarily expensive compute choice (e.g., using All-Purpose for a scheduled job).

2 · Check your understanding