Prepare and test your skills

Prepare and test your skills

Free sample question with answers and detailed explanation below. The Cloud Digital Leader practice bank has an additional 3,070 practice questions.Try 10 free realistic questions: instant score, no credit card, and no account required.
An organization is using TensorFlow to build and train a large-scale machine learning model. The engineering team needs to provision and manage Cloud Tensor Processing Units (TPUs) using Compute Engine managed instance groups (MIGs) to support the following workload requirements:
Which Cloud TPU configuration should the organization implement to meet these requirements?
Cloud Tensor Processing Units (TPUs) are Google's custom-developed application-specific integrated circuits (ASICs) engineered specifically to accelerate artificial intelligence and machine learning workloads, such as deep learning models built with TensorFlow. When provisioned through Compute Engine managed instance groups (MIGs) as multi-host TPU slices, multiple TPU virtual machines (VMs) are interconnected to operate cohesively as a single processing unit.
Unlike single-host configurations that isolate workloads onto independent machines, a multi-host Cloud TPU slice managed via a Compute Engine MIG coordinates multiple interconnected VMs over high-speed networks. This ensures that a massive, unified TensorFlow training job can distribute its matrix calculations efficiently across the entire slice while benefiting from automated Compute Engine instance group management.
A Compute Engine managed instance group (MIG) configured with single-host Cloud TPU VMs provisions multiple standalone TPU virtual machines that are managed under a single group policy. While they share lifecycle management rules within Compute Engine, each single-host TPU VM operates independently rather than being interconnected into a unified multi-host compute slice.
This configuration would be the right choice if the organization had already finished training its TensorFlow model and needed to deploy it to run multiple, independent inference workloads concurrently across separate TPU hosts.
A common misconception is that placing any TPU VMs inside a managed instance group (MIG) automatically interconnects them for distributed training. In reality, administrators must specifically choose between single-host VMs (for independent tasks like inference) and multi-host TPU slices (for unified, large-scale training workloads).
This option describes provisioning standalone TPU virtual machines (VMs) without interconnected slice scaling, while attempting to use the hardware for workloads dominated by frequent branching logic, element-wise algebra, and high-precision arithmetic calculations.
Workloads requiring frequent branching, element-wise operations, or high-precision arithmetic would be appropriate if the team were provisioning general-purpose CPUs or specialized high-precision GPUs on Compute Engine for traditional scientific simulations, rather than using Cloud TPUs for matrix-heavy TensorFlow deep learning.
Understanding when to use Cloud TPUs is essential when designing AI solutions on Google Cloud. TPUs excel when training large TensorFlow, JAX, or PyTorch models dominated by matrix computations, large batch sizes, and ultra-large embeddings, whereas workloads with custom operations in the main training loop or complex branching should run on alternative compute options.
The Cloud TPU API—along with its associated Google Cloud CLI commands and Cloud Client Libraries for Cloud TPU—represents Google Cloud's older, standalone interface originally created for provisioning and interacting with TPU hardware resources outside of standard Compute Engine virtual machine workflows.
This option would only be encountered if a team were auditing an older, legacy machine learning environment that had not yet been modernized. Even in that scenario, best practices require migrating away from the legacy Cloud TPU API to Compute Engine or GKE.
To streamline how TensorFlow and other machine learning frameworks integrate with cloud infrastructure, Google Cloud consolidated TPU provisioning into Compute Engine (for full VM and slice lifecycle management) and GKE (for containerized, multi-tenant orchestration). This allows teams to manage specialized AI hardware using the same managed instance groups and Kubernetes node pools they use for standard cloud compute resources.
Answer them yourself and get your score the moment you finish. No account needed.
Full access includes
Keep the momentum going with these hand-picked practice scenarios
Last updated