Cloud Digital Leader Google Cloud Products and Services Practice Question

Answer analysis

Option-by-option breakdown

For each option: why learners choose it and why it is or isn't the right answer here.

• ✓

  Vertex AI

  Why this is correct

  Vertex AI is a fully managed machine learning platform that supports custom container training, allowing you to bring your own model code and dependencies while leveraging managed GPU and TPU clusters. It provides automatic scaling, hyperparameter tuning, and integrated MLOps tools like model versioning, monitoring, and Vertex Pipelines, which drastically reduce operational overhead compared to raw compute. This makes it the ideal choice for running custom ML models at scale without manually provisioning infrastructure.

• ✗

  Cloud Run

  Why it's wrong here

  Cloud Run is a serverless compute platform for running stateless containers that respond to web requests or events, but it does not currently support GPU acceleration, which is critical for most machine learning training workloads. Additionally, Cloud Run has a maximum request timeout of 60 minutes, which is insufficient for long-running training jobs that can take hours, and it lacks native support for distributed training across multiple workers. Therefore, it is unsuitable for this custom ML training use case.

• ✗

  Compute Engine

  Why it's wrong here

  Compute Engine offers custom virtual machines with GPU support, but it requires you to manually provision, configure, and manage the VM instances, including installing GPU drivers, managing software updates, and handling hardware failures. You also need to manually implement autoscaling and fault tolerance, often using instance groups, and you pay for idle capacity unless you carefully manage lifecycle policies. This operational burden makes it less efficient and more error-prone than a managed ML platform like Vertex AI.

• ✗

  Cloud Functions

  Why it's wrong here

  Cloud Functions is a serverless function-as-a-service platform designed for event-driven, short-lived tasks, with maximum execution times usually around 9 minutes and no support for GPU accelerators. It is not designed for compute-intensive workloads like ML training, as it lacks persistent state, distributed training capabilities, and the ability to handle custom containers with complex dependencies. Thus, it is completely inappropriate for training a custom machine learning model.

Go deeper

Related to this question