- A
Switch to a larger GPU such as V100 or A100 to increase per-replica throughput.
Switching to a larger GPU (e.g., V100 or A100) increases per-replica compute capacity, directly addressing the GPU saturation and reducing 503 errors. This is the most effective fix.
- B
Implement request batching in the custom container to improve GPU utilization efficiency.
Why wrong: Request batching can improve GPU utilization efficiency, but with GPU already at 90%+, batching gains are limited. The primary issue is insufficient GPU compute capacity, not underutilization.
- C
Enable model parallelism across multiple GPUs within each replica.
Why wrong: Model parallelism is typically used for models that exceed GPU memory. With a 2 GB model, it adds overhead without benefit. The bottleneck is compute, not memory.
- D
Use a high-memory machine type like n1-highmem-16 to reduce memory pressure.
Why wrong: Increasing memory (n1-highmem-16) does not help because memory usage is low (<50%). The bottleneck is GPU compute, not memory capacity or caching.
GPU Utilization Bottleneck on Vertex AI
This PMLE practice question tests your understanding of gpu bottleneck. The scenario asks you to isolate a root cause — eliminate options that address a different problem before choosing. A key principle to apply: gPU Bottleneck. Once you have made your selection, read the full explanation to reinforce the concept and understand why each distractor is designed to mislead on exam day.
Your organization has a large production system that uses Vertex AI Prediction for an NLP model with a 2 GB memory footprint. The endpoint is configured with 5 replicas, each using an n1-standard-4 with a single T4 GPU. Recently, you observed an increase in 503 errors during peak hours. Cloud Monitoring shows that GPU utilization is consistently above 90% across all replicas, while CPU and memory are below 50%. You have already increased the max replicas to 10, but the errors persist because the increased replicas also become saturated. What should you do to resolve the issue?
Quick Answer
The answer is to use a high-memory machine type like n1-highmem-16 to reduce memory pressure. This resolves the GPU utilization bottleneck on Vertex AI because the NLP model’s 2 GB memory footprint is causing excessive GPU idle time as the system swaps data between RAM and GPU memory, saturating the GPU at over 90% while CPU and memory remain below 50%. On the Google Professional Machine Learning Engineer exam, this scenario tests your understanding that GPU utilization bottlenecks often stem from insufficient host memory, not GPU compute power—a common trap where candidates mistakenly scale replicas or CPU cores. The key insight is that Vertex AI Prediction endpoints require balanced resources; here, adding memory per replica allows the GPU to process batches without stalling. Memory tip: think “GPU hungry, memory shy”—when GPU is pegged but CPU is low, feed it more RAM, not more replicas.
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
Switch to a larger GPU such as V100 or A100 to increase per-replica throughput.
GPU utilization is above 90% across all replicas, indicating the GPU is the bottleneck. Increasing replicas does not help because each new replica also saturates its GPU. The most direct solution is to increase per-replica throughput by using a more powerful GPU (e.g., V100 or A100), which can process more work per unit time. Option A addresses the root cause—GPU capacity. Option D (high-memory machine) does not help because memory is not the constraint (CPU/memory < 50%). Batching (B) is likely already in use but limited by GPU compute capacity. Model parallelism (C) is unnecessary for a 2 GB model.
Key principle: GPU Bottleneck
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✓
Switch to a larger GPU such as V100 or A100 to increase per-replica throughput.
Why this is correct
Switching to a larger GPU (e.g., V100 or A100) increases per-replica compute capacity, directly addressing the GPU saturation and reducing 503 errors. This is the most effective fix.
Related concept
GPU Bottleneck
- ✗
Implement request batching in the custom container to improve GPU utilization efficiency.
Why it's wrong here
Request batching can improve GPU utilization efficiency, but with GPU already at 90%+, batching gains are limited. The primary issue is insufficient GPU compute capacity, not underutilization.
- ✗
Enable model parallelism across multiple GPUs within each replica.
Why it's wrong here
Model parallelism is typically used for models that exceed GPU memory. With a 2 GB model, it adds overhead without benefit. The bottleneck is compute, not memory.
- ✗
Use a high-memory machine type like n1-highmem-16 to reduce memory pressure.
Why it's wrong here
Increasing memory (n1-highmem-16) does not help because memory usage is low (<50%). The bottleneck is GPU compute, not memory capacity or caching.
Common exam traps
Common exam trap: answer the scenario, not the keyword
Candidates may assume that adding more replicas or memory will solve throughput issues, but when the GPU is the bottleneck, the only effective remedy is to upgrade the GPU itself.
Detailed technical explanation
How to think about this question
Treat this as a scenario question. Identify the problem, the constraint, and the best action. Then compare each option against those facts.
KKey Concepts to Remember
- GPU Bottleneck
- Vertical Scaling
- Resource Saturation
TExam Day Tips
- Watch for words such as best, first, most likely and least administrative effort.
- Review why wrong options are wrong, not only why the correct option is correct.
Key takeaway
GPU Bottleneck
Real-world example
How this comes up in practice
An e-commerce site experiences heavy traffic on Black Friday and near-zero traffic during off-peak weeks. Rather than provisioning permanent large VMs, the team uses auto-scaling groups that add capacity automatically under load and reduce it overnight. Questions like this test whether you understand elasticity, availability zones, and cloud compute scaling patterns.
Visual reference
What to study next
Got this wrong? Here's your next step.
Review gPU Bottleneck, then practise related PMLE questions on the same topic to reinforce the concept.
Related practice questions
Related PMLE practice-question pages
Use these pages to review the topic behind this question. This is how one missed question becomes focused revision.
Automating and Orchestrating ML Pipelines practice questions
Practise PMLE questions linked to Automating and Orchestrating ML Pipelines.
Collaborating Within and Across Teams to Manage Data and Models practice questions
Practise PMLE questions linked to Collaborating Within and Across Teams to Manage Data and Models.
Serving and Scaling Models practice questions
Practise PMLE questions linked to Serving and Scaling Models.
Monitoring ML Solutions practice questions
Practise PMLE questions linked to Monitoring ML Solutions.
Architecting Low-Code ML Solutions practice questions
Practise PMLE questions linked to Architecting Low-Code ML Solutions.
Scaling Prototypes into ML Models practice questions
Practise PMLE questions linked to Scaling Prototypes into ML Models.
Collaborating to manage data and models practice questions
Practise PMLE questions linked to Collaborating to manage data and models.
Solving business challenges with ML practice questions
Practise PMLE questions linked to Solving business challenges with ML.
PMLE fundamentals practice questions
Practise PMLE questions linked to PMLE fundamentals.
PMLE scenario practice questions
Practise PMLE questions linked to PMLE scenario.
PMLE troubleshooting practice questions
Practise PMLE questions linked to PMLE troubleshooting.
Practice this exam
Start a free PMLE practice session
Short sessions build daily habit. Longer sessions build exam-day stamina. Try a timed session to simulate real conditions.
FAQ
Questions learners often ask
What does this PMLE question test?
GPU Bottleneck
What is the correct answer to this question?
The correct answer is: Switch to a larger GPU such as V100 or A100 to increase per-replica throughput. — GPU utilization is above 90% across all replicas, indicating the GPU is the bottleneck. Increasing replicas does not help because each new replica also saturates its GPU. The most direct solution is to increase per-replica throughput by using a more powerful GPU (e.g., V100 or A100), which can process more work per unit time. Option A addresses the root cause—GPU capacity. Option D (high-memory machine) does not help because memory is not the constraint (CPU/memory < 50%). Batching (B) is likely already in use but limited by GPU compute capacity. Model parallelism (C) is unnecessary for a 2 GB model.
What should I do if I get this PMLE question wrong?
Review gPU Bottleneck, then practise related PMLE questions on the same topic to reinforce the concept.
What is the key concept behind this question?
GPU Bottleneck
About these practice questions
Courseiva creates original exam-style practice questions with explanations and wrong-answer analysis. It does not publish real exam questions, exam dumps, or protected exam content. Learn why practice questions differ from exam dumps →
Keep practising
More PMLE practice questions
- A travel booking company has a real-time recommendation system that suggests hotels and flights to users. The model is s…
- A global retail company uses Vertex AI Recommendations to provide product recommendations on their website. They have a…
- Your team is developing a machine learning model for real-time fraud detection. The training pipeline runs on Vertex AI…
- A healthcare organization is building a machine learning model to predict patient readmission risk. They have sensitive…
- You are an ML engineer at a global e-commerce company. Your team has developed a deep learning model for product recomme…
- A financial services company uses Vertex AI AutoML Tables to build a credit risk model. The dataset contains 500,000 row…
Last reviewed: Jun 24, 2026
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.
Question Discussion
Share a tip, memory trick, or ask about the reasoning behind this question. Do not post real exam questions, leaked content, braindumps, or copyrighted exam material. Comments are moderated and may be removed without notice.
Sign in to join the discussion.