AI-300 Practice Question 1: Training Compute Target
Microsoft Machine Learning Operations Engineer Associate (AI-300) · Test 1 · question 1 of 10
A data scientist currently develops notebooks on an Azure Machine Learning compute instance. The team now needs to run a nightly training job over a much larger dataset. The job takes about two hours, and nobody should be charged for compute during the other 22 hours of the day. Which compute target should you configure for the nightly job?
Show answer and explanation
Answer:
An Azure Machine Learning compute cluster is managed compute that autoscales each time you submit a job. The documented way to avoid paying for idle compute on a cluster is to set the minimum number of nodes to 0, so the cluster scales back to nothing between nightly runs and scales out again when the next job is submitted (option B).
- A is wrong because Azure Databricks is an unmanaged compute target you attach and maintain separately. It adds a second platform and its own cluster lifecycle to manage, which is unnecessary for a plain Azure Machine Learning training job. - C is wrong because a compute instance is a single-node target that does not autoscale per job. Idle shutdown helps, but you still pay for disk, public IP, and load balancer, and a bigger VM size makes the idle hours more expensive rather than free. - D is wrong because an attached remote VM is unmanaged: it doesn't autoscale, so you either leave it running and pay for 24 hours or take on starting and stopping it yourself.
Reference: Understand compute targets - Azure Machine Learning