- A
Compile the model using TensorFlow XLA to optimize for single GPU execution.
Why wrong: XLA improves performance but cannot make a model that doesn't fit on one GPU fit; parallelism is required.
- B
Deploy the model across multiple endpoints and use a load balancer to send requests to different parts of the model.
Why wrong: This would require splitting the model across nodes, adding network overhead and increasing latency beyond 500ms.
- C
Use Vertex AI Prediction as a service for LLMs, which automatically handles hardware selection.
Why wrong: Vertex AI Prediction may not support custom LLMs with specific parallelism needs; it's for managed models.
- D
Use a machine type with multiple GPUs and configure the container to use tensor parallelism.
Leveraging multiple GPUs on one node via model parallelism (e.g., tensor parallelism) is the standard approach to fit large models and meet latency.
How to Achieve 500ms Latency for Large Models Using Tensor Parallelism
This PDE practice question tests your understanding of operationalizing machine learning models. Read the scenario carefully and evaluate each option against the stated constraints before committing to an answer. After answering, compare your reasoning against the explanation and wrong-answer breakdown below. 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.
You are designing a system to serve predictions from a large language model (LLM) with a latency SLO of 500ms. The model does not fit on a single GPU and requires model parallelism. You are considering using Vertex AI Endpoints with a custom container. What additional setup is required to achieve the latency target?
Quick Answer
The answer is to use a machine type with multiple GPUs and configure the container to use tensor parallelism. This approach is correct because tensor parallelism splits the large language model’s layers across multiple GPUs on a single machine, allowing simultaneous computation that dramatically reduces inference time to meet a 500ms latency SLO. Unlike sharding across separate endpoints, which introduces network latency, tensor parallelism keeps communication within the high-speed GPU interconnect, making it ideal for model parallelism for LLMs on Vertex AI. On the Google Professional Data Engineer exam, this tests your understanding of Vertex AI Endpoints’ custom container capabilities and the trade-offs between intra-machine parallelism and distributed serving. A common trap is assuming that simply using more machines or TPUs is required, but the key is multi-GPU configuration within one node. Memory tip: think “tensor splits, latency fits” to remember that splitting tensors across GPUs on the same machine is the fastest path to sub-second latency.
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
Use a machine type with multiple GPUs and configure the container to use tensor parallelism.
Option D is correct because the model does not fit on a single GPU and requires model parallelism. Using a machine type with multiple GPUs and configuring the container to use tensor parallelism allows the model to be split across GPUs within a single instance, enabling efficient parallel computation to meet the 500ms latency SLO. Tensor parallelism distributes individual tensor operations across GPUs, reducing communication overhead compared to pipeline parallelism and is a standard approach for large models on multi-GPU instances in Vertex AI.
Key principle: Answer the scenario, not the keyword: identify the specific constraint before choosing the most familiar-sounding option.
Answer analysis
Option-by-option breakdown
For each option: why learners choose it and why it is or isn't the right answer here.
- ✗
Compile the model using TensorFlow XLA to optimize for single GPU execution.
Why it's wrong here
XLA improves performance but cannot make a model that doesn't fit on one GPU fit; parallelism is required.
- ✗
Deploy the model across multiple endpoints and use a load balancer to send requests to different parts of the model.
Why it's wrong here
This would require splitting the model across nodes, adding network overhead and increasing latency beyond 500ms.
- ✗
Use Vertex AI Prediction as a service for LLMs, which automatically handles hardware selection.
Why it's wrong here
Vertex AI Prediction may not support custom LLMs with specific parallelism needs; it's for managed models.
- ✓
Use a machine type with multiple GPUs and configure the container to use tensor parallelism.
Why this is correct
Leveraging multiple GPUs on one node via model parallelism (e.g., tensor parallelism) is the standard approach to fit large models and meet latency.
Related concept
Read the scenario before looking for a memorised answer.
Common exam traps
Common exam trap: answer the scenario, not the keyword
Candidates often incorrectly assume that distributing inference across multiple Vertex AI endpoints with a load balancer achieves model parallelism. However, this approach introduces network latency and cannot match the low-latency inter-GPU communication required for tensor parallelism within a single multi-GPU instance.
Detailed technical explanation
How to think about this question
Tensor parallelism in Vertex AI custom containers typically leverages frameworks like NVIDIA Megatron-LM or DeepSpeed, which shard model parameters across GPUs and synchronize gradients via NCCL. The key challenge is ensuring that the container's entrypoint correctly initializes the distributed runtime (e.g., via torch.distributed or Horovod) and that the Vertex AI endpoint is configured with the appropriate machine type (e.g., a2-highgpu-8g) to expose multiple GPUs. In practice, achieving sub-500ms latency may also require optimizing the model's batch size and using GPU direct RDMA to minimize inter-GPU communication latency.
KKey Concepts to Remember
- Read the scenario before looking for a memorised answer.
- Find the constraint that changes the correct option.
- Eliminate answers that are true in general but not in this case.
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
Answer the scenario, not the keyword: identify the specific constraint before choosing the most familiar-sounding option.
Real-world example
How this comes up in practice
A cloud solutions architect for a retail company is evaluating services for a new workload. The correct answer here reflects best practice for the specific scenario described — not a general cloud recommendation. Answer the scenario, not the keyword: identify the specific constraint before choosing the most familiar-sounding option. Cloud exam questions reward reading the constraint carefully: the same technology can be right or wrong depending on the use case.
What to study next
Got this wrong? Here's your next step.
Identify which exam domain this question belongs to, review the core concept, then practise similar questions from the same domain.
- →
Operationalizing machine learning models — study guide chapter
Learn the concepts, then practise the questions
- →
Operationalizing machine learning models practice questions
Targeted practice on this topic area only
- →
All PDE questions
1,000 questions across all exam domains
- →
Google Professional Data Engineer study guide
Full concept coverage aligned to exam objectives
- →
PDE practice test guide
How to use practice tests most effectively before exam day
Related practice questions
Related PDE practice-question pages
Use these pages to review the topic behind this question. This is how one missed question becomes focused revision.
Designing Data Processing Systems practice questions
Practise PDE questions linked to Designing Data Processing Systems.
Ingesting and Processing the Data practice questions
Practise PDE questions linked to Ingesting and Processing the Data.
Storing the Data practice questions
Practise PDE questions linked to Storing the Data.
Preparing and Using Data for Analysis practice questions
Practise PDE questions linked to Preparing and Using Data for Analysis.
Maintaining and Automating Data Workloads practice questions
Practise PDE questions linked to Maintaining and Automating Data Workloads.
Building and operationalizing data processing systems practice questions
Practise PDE questions linked to Building and operationalizing data processing systems.
Operationalizing machine learning models practice questions
Practise PDE questions linked to Operationalizing machine learning models.
Ensuring solution quality practice questions
Practise PDE questions linked to Ensuring solution quality.
PDE fundamentals practice questions
Practise PDE questions linked to PDE fundamentals.
PDE scenario practice questions
Practise PDE questions linked to PDE scenario.
PDE troubleshooting practice questions
Practise PDE questions linked to PDE troubleshooting.
Practice this exam
Start a free PDE 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 PDE question test?
Operationalizing machine learning models — This question tests Operationalizing machine learning models — Read the scenario before looking for a memorised answer..
What is the correct answer to this question?
The correct answer is: Use a machine type with multiple GPUs and configure the container to use tensor parallelism. — Option D is correct because the model does not fit on a single GPU and requires model parallelism. Using a machine type with multiple GPUs and configuring the container to use tensor parallelism allows the model to be split across GPUs within a single instance, enabling efficient parallel computation to meet the 500ms latency SLO. Tensor parallelism distributes individual tensor operations across GPUs, reducing communication overhead compared to pipeline parallelism and is a standard approach for large models on multi-GPU instances in Vertex AI.
What should I do if I get this PDE question wrong?
Identify which exam domain this question belongs to, review the core concept, then practise similar questions from the same domain.
What is the key concept behind this question?
Read the scenario before looking for a memorised answer.
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 PDE practice questions
- A company wants to process large CSV files stored in Cloud Storage and load them into BigQuery. The files are generated…
- A company runs a Dataflow streaming pipeline that reads from Cloud Pub/Sub and writes to BigQuery. The pipeline uses a s…
- A company uses Cloud Dataproc for ephemeral clusters to run batch jobs. They want to ensure job reliability and data qua…
- Your company uses Vertex AI Pipelines to automate model retraining. The pipeline has three steps: data extraction from B…
- A company wants to use BigQuery to query data stored in Parquet files in Cloud Storage without loading the data into Big…
- A company has deployed a machine learning model to AI Platform Prediction. The model uses a custom container with a Tens…
Last reviewed: Jul 4, 2026
This PDE 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 PDE 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.