GCP Professional Cloud Architect Practice Question

Your company runs three independent workloads in Google Cloud:

Workload A: A once-per-month ETL backfill that spins up thousands of stateless workers for about 12 hours. Tasks checkpoint progress every few minutes and can restart automatically if an instance disappears.
Workload B: A low-latency pricing service that must run on exactly 10 vCPU and 32 GB RAM per replica to align with a fixed software license model.
Workload C: An ML research team that repeatedly trains large transformer models needing eight NVIDIA A100 GPUs per run and wants the shortest possible training time.

Which Compute Engine configuration is the most cost-effective match for these workloads while satisfying their technical constraints?

A → N2 standard instances with a 3-year committed-use discount, B → predefined n2-standard-16 machines, C → Cloud TPUs v4 pods
A → Reserved sole-tenant nodes, B → M2-ultramem-208 machines, C → N1-standard-32 instances with P100 GPUs
A → Autoscaled Cloud Run jobs, B → E2-standard-8 instances behind an internal load balancer, C → T2D-standard-32 instances with attached NVIDIA T4 GPUs
A → Spot VMs in a managed instance group, B → custom 10 vCPU/32 GB machine type, C → A2-highgpu (8×A100) instances

Report Issue

Answer Description

Workload A benefits most from Spot VMs because it is fault-tolerant, short-lived, and runs only once a month; accepting preemption cuts instance cost by up to 91% without affecting results.
Workload B needs a precise vCPU-to-memory ratio to avoid over-provisioning licensed cores; custom machine types let you define a 10 vCPU/32 GB shape so you pay only for what is required while meeting latency objectives.
Workload C needs multiple A100 GPUs and high GPU throughput; the A2-highgpu family provides up to 8 A100 GPUs with optimized CPU-to-GPU interconnect and is the recommended choice for large-scale training.
Alternative options either cost more or do not meet technical needs: sustained-use or committed-use standard N2 instances for Workload A miss the largest savings; predefined N1/N2 shapes for Workload B waste vCPUs or memory; TPUs do not run GPU-specific code and smaller GPU VMs would prolong training. Therefore, mapping Spot VMs for A, custom machine types for B, and A2-highgpu VMs for C is correct.

Ask Bash

Bash is our AI bot, trained to help you pass your exam. AI Generated Content may display inaccurate information, always double-check anything important.