MLA-C01 Question 265
Select 2You are deploying a machine learning model in a production environment using AWS SageMaker. The model requires GPU instances for inference due to its computationally intensive nature. Meanwhile, you also need to set up a testing environment to evaluate the model on a smaller dataset before deployment. How should you provision the compute resources for both environments to optimize cost and performance?
- A
Use an instance type with GPU support, such as ml.p3.2xlarge, in both production and testing environments to ensure consistency.
- B
Use an instance type with GPU support, such as ml.p3.2xlarge, in the production environment and a CPU instance, such as ml.m5.large, in the testing environment.
- C
Use a multi-GPU instance, such as ml.g4dn.12xlarge, in both production and testing environments for maximum performance.
- D
Use a CPU instance, such as ml.m5.large, for both production and testing environments to reduce costs.
- E
Use an instance type with GPU support in the production environment and leverage SageMaker local mode with a CPU for testing.
Show answer and explanation
Correct answers: B, E
Explanation
In production environments, it is important to select instance types that offer the necessary performance for the workload; in this case, GPU instances are required for inference. However, a testing environment typically works with a smaller dataset and does not need the same level of computational power. Using a CPU instance or SageMaker local mode for testing reduces costs while still enabling effective evaluation of the model. Combining GPU-backed instances in production and cost-effective methods for testing achieves an optimal balance between cost and performance.
- A. Incorrect.
While using GPU instances for both production and testing ensures consistency, it is not cost-efficient for the testing environment, which typically doesn’t require intensive resources.
- B. Correct.
This is a valid approach, as it balances cost and performance by using a GPU instance for the production environment where inference speed matters and a CPU instance for the testing environment where the computational demand is lower.
- C. Incorrect.
Using a multi-GPU instance for both environments is unnecessary and costly, especially for a testing environment that doesn’t require such high computational power.
- D. Incorrect.
Using CPU instances for both environments does reduce costs but may lead to suboptimal performance in the production environment where GPU acceleration is required for intensive inference tasks.
- E. Correct.
This is a valid approach, as SageMaker local mode can simulate the model behavior on a smaller dataset using a CPU, significantly reducing costs and still enabling efficient testing. Production, on the other hand, requires GPU resources for optimal inference.