easyMultiple Choice
Autoscaling GPU Models on Vertex AI Prediction — Ensuring GPU Quota
A company has deployed a TensorFlow model on Vertex AI Prediction for real-time inference. They notice that during peak hours, the prediction latency increases significantly, and some requests time out. The model requires GPU acceleration. Which action should they take to reduce latency and avoid timeouts?
Quick Answer
The correct answer is to enable autoscaling with min replicas set to the base load and max replicas set to handle peak load, while ensuring GPU quota is sufficient. This approach dynamically adjusts compute resources in response to traffic, preventing latency spikes and timeouts during surges without wasting resources during idle periods. On the Google Professional Machine Learning Engineer exam, this scenario tests your understanding of Vertex AI Prediction’s autoscaling behavior and the critical role of GPU quota—a common trap is assuming static scaling or CPU upgrades solve GPU-bound bottlenecks. Remember, autoscaling relies on quota headroom; without pre-allocated GPU quota, the service cannot spin up additional replicas under load. A simple memory tip: “Min for base, max for burst, quota for worst.”
Answer choices
Why each option matters
Answer the question above first, then reveal the full breakdown to understand why each option is right or wrong.
Correct answer & explanation
✓
Enable autoscaling with min replicas set to the base load and max replicas set to handle peak load, and ensure GPU quota is sufficient.
Enabling autoscaling with appropriate min and max replicas allows the endpoint to dynamically scale up during peak traffic and scale down during low traffic, ensuring sufficient GPU resources to handle the load without manual intervention. Ensuring adequate GPU quota is also critical to prevent resource exhaustion. Option B is wrong because switching to a larger machine type with more vCPUs does not address GPU-bound inference latency; the bottleneck is GPU acceleration, not CPU. Option C is wrong because statically increasing replicas leads to over-provisioning and waste during off-peak hours, and cannot react quickly to sudden traffic spikes. Option D is wrong because Cloud Functions are serverless and do not support GPU acceleration; they add additional latency and are not suitable for real-time GPU inference.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Enable autoscaling with min replicas set to the base load and max replicas set to handle peak load, and ensure GPU quota is sufficient.
Why this is correct
Enabling autoscaling with appropriate min and max replicas allows the endpoint to dynamically scale up during peak traffic and scale down during low traffic, ensuring sufficient GPU resources to handle the load without manual intervention. Ensuring adequate GPU quota is also critical to prevent resource exhaustion.
- ✗
Switch to a larger machine type with more vCPUs.
Why it's wrong here
Adding vCPUs does not accelerate GPU-bound TensorFlow inference, and the stem states the model requires GPU acceleration, so latency persists. It is tempting because larger machine types often relieve CPU-bound serving bottlenecks, and that would be the right move for a CPU-limited preprocessing or batching stage.
- ✗
Increase the number of replicas in the Vertex AI Prediction endpoint statically to handle peak load.
Why it's wrong here
Static replica increases add capacity for concurrent requests but each replica still queues behind GPU-bound inference, so per-request latency during peaks remains. It is tempting because horizontal scaling is the standard answer for throughput, and it would be correct if the endpoint were CPU-bound or simply saturated on request volume.
- ✗
Use Cloud Functions to invoke the model asynchronously.
Why it's wrong here
Cloud Functions invoking the model asynchronously removes the caller's wait but does not reduce inference latency, and Vertex AI real-time endpoints still time out under load. It is tempting because asynchronous invocation suits batch or fire-and-forget workloads, where the client does not need an immediate prediction.
Quick reference
Cloud Service Model Comparison
| Model | You Manage | Provider Manages | Examples |
|---|---|---|---|
| IaaS | OS, runtime, apps, data | Hardware, hypervisor, networking | EC2, Azure VMs, GCP Compute Engine |
| PaaS | Apps and data | OS, runtime, middleware, hardware | Elastic Beanstalk, Azure App Service |
| SaaS | Data and settings only | Everything else | Microsoft 365, Salesforce, Workday |
| FaaS / Serverless | Function code only | Infra, scaling, runtime | Lambda, Azure Functions, Cloud Run |
| CaaS | Containers and apps | Kubernetes, OS, hardware | EKS, AKS, GKE |
Go deeper
Related to this question
About these practice questions
Courseiva writes every PMLE question from scratch — 775 in total, each with an explanation and a wrong-answer breakdown. None are copied from real exams or dumps. Learn why practice questions differ from exam dumps →
Same concept, more angles
1 more way this is tested on PMLE
These questions test the same concept from different angles. Work through them to make sure you can recognise it however the exam phrases it.
Variation 1. A company is deploying a model for online predictions on Vertex AI. They want to minimize latency while also handling traffic spikes. Which TWO configurations should they choose?
medium- ✓ A.Use GPU machine type
- ✓ B.Enable autoscaling with min replicas=1
- C.Disable autoscaling and use manual scaling
- D.Use CPU machine type with more memory
- E.Set a fixed number of replicas equal to peak load
Why A: Option A is correct because deploying the model on a GPU machine type provides hardware acceleration that reduces inference latency for compute-intensive models, which directly addresses the requirement to minimize latency for online predictions. Option B is correct because enabling autoscaling with min replicas=1 keeps at least one replica always available to serve requests (avoiding cold-start latency) while allowing Vertex AI to automatically add replicas when traffic spikes, satisfying the need to handle traffic surges. Option C is incorrect because disabling autoscaling and using manual scaling cannot react to sudden traffic spikes and risks either under-provisioning or over-provisioning. Option D is incorrect because a CPU machine type with more memory does not provide the same low-latency acceleration as a GPU and does not address traffic spikes. Option E is incorrect because a fixed number of replicas sized to peak load wastes resources during normal traffic and still cannot adapt if demand exceeds the assumed peak.
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This PMLE practice question is part of Courseiva's free Google Cloud certification practice question bank. Courseiva provides original exam-style practice questions with explanations, topic-based practice, mock exams, readiness tracking, and study analytics to help learners prepare for the PMLE exam.