MLA-C01 Question 264
Select 2You are deploying a machine learning model in a production environment. The model requires high inference speed, and the workload involves processing large batches of data using deep learning algorithms. In a separate test environment, you need to evaluate the model's performance with smaller datasets and minimal cost. What is the best approach to provision compute resources for these environments?
- A
Use AWS EC2 instances with GPU support (e.g., P3 or G5 instances) for the production workload and CPU-only instances (e.g., T3 instances) for the test environment.
- B
Use AWS SageMaker endpoints with multi-model deployment for the production workload and SageMaker notebook instances for the test environment.
- C
Use AWS Lambda for both production and test environments to minimize costs.
- D
Use AWS EC2 Spot instances with GPU support for the production workload and On-Demand EC2 instances with CPU support for the test environment.
- E
Use AWS SageMaker endpoints with GPU acceleration for production and SageMaker Processing Jobs for the test environment.
Show answer and explanation
Correct answers: A, E
Explanation
When provisioning compute resources for production and test environments, it is essential to prioritize performance and cost-efficiency. GPU-enabled resources like EC2 P3 or G5 instances or SageMaker endpoints with GPU acceleration are optimal for high-performance deep learning inference in production. For testing environments, cost-effective solutions like CPU-based instances (e.g., T3) or SageMaker Processing Jobs are suitable for evaluating models with smaller datasets. This approach balances performance needs in production and cost-efficiency in testing.
- A. Correct.
This is correct because GPU-supported EC2 instances (e.g., P3 or G5) are ideal for high-performance deep learning inference in production, while CPU-only instances (e.g., T3) are cost-effective for testing with smaller datasets.
- B. Incorrect.
This is incorrect because multi-model deployment with SageMaker endpoints is not specifically required for this scenario, and SageMaker notebook instances are typically used for development, not testing production-like workloads.
- C. Incorrect.
This is incorrect because AWS Lambda is not suited for high-performance inference or GPU-based processing required in production for deep learning workloads.
- D. Incorrect.
This is partially correct but not ideal because Spot instances may not guarantee the availability required for production workloads, and On-Demand EC2 instances can be more expensive than necessary for test environments.
- E. Correct.
This is correct because SageMaker endpoints with GPU acceleration provide a managed solution for high-performance inference in production, while SageMaker Processing Jobs are cost-effective for testing with smaller datasets.