AI-102 Implement generative AI solutions Practice Question
You are deploying a generative AI model using Azure Machine Learning. The model is a large language model that requires GPU compute. You need to minimize costs while ensuring the model is always available for inference. Which compute option should you choose?
⚠ Common exam trap
Microsoft often tests the distinction between 'always available' (requiring a persistent compute instance) and 'cost-optimized' (allowing scale-to-zero), leading candidates to mistakenly choose serverless GPU compute because it sounds cheaper, but it fails the availability requirement.
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
✓
Managed online endpoint with a GPU VM
A managed online endpoint with a GPU VM is the correct choice because it provisions a dedicated GPU instance that remains always active, ensuring the model is available for real-time inference at any time. This option balances cost and availability by allowing you to choose a lower-cost GPU SKU (e.g., NCas_v4) while avoiding the cold-start latency of serverless or batch options. The managed endpoint also handles auto-scaling and load balancing, but for constant availability, a fixed minimum instance count is required.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Managed online endpoint with a GPU VM
Why this is correct
Managed online endpoints provide real-time inference with GPU, and autoscaling can help cost while keeping availability.
- ✗
Batch endpoint with GPU
Why it's wrong here
A batch endpoint processes inference jobs asynchronously on a schedule, not in real time. The scenario requires the model to be always available for inference, which demands a persistent, always-on compute target such as an online endpoint. Batch endpoints are tempting because they reduce cost by scaling GPU resources to zero between job runs, making them ideal for offline or periodic inference workloads where latency is not critical.
- ✗
Serverless GPU compute
Why it's wrong here
Serverless may have cold start delays and is not always immediately available.
- ✗
CPU-based inference
Why it's wrong here
CPU is too slow for large language models.
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 AI-102 question from scratch — 945 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 →
JA
Written by Johnson Ajibi, MSc IT Security
Senior Network & Security Engineer · founder of Courseiva
This AI-102 practice question is part of Courseiva's free Microsoft 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 AI-102 exam.