MLS-C01 Question 326
Select 2You are building a machine learning model in AWS SageMaker that requires GPU acceleration during training. Once training is complete, you need to deploy the model with minimal cost while maintaining low latency for real-time inference. Which combination of instance types and configurations should you choose to meet these requirements?
- A
Use a p3.2xlarge instance for training and an ml.m5.large instance for deployment.
- B
Use a g4dn.xlarge instance for training and an ml.c5.large instance for deployment.
- C
Use a p3.2xlarge instance for both training and deployment.
- D
Use an ml.m5.large instance for training and a p3.2xlarge instance for deployment.
- E
Use a g4dn.xlarge instance for training and an ml.g4dn.xlarge instance for deployment.
Show answer and explanation
Correct answers: A, B
Explanation
To optimize both cost and performance, you should select a GPU-enabled instance (like p3 or g4dn series) for training, as they are designed for computationally intensive machine learning tasks. For deployment, choose a cost-efficient CPU-based instance (like m5 or c5 series) unless GPU acceleration is explicitly required for inference. This approach minimizes costs while ensuring the training and deployment needs are met.
- A. Correct.
Correct: The p3.2xlarge instance provides GPU acceleration for training, and the ml.m5.large instance is cost-effective for deployment, as it does not require GPUs.
- B. Correct.
Correct: The g4dn.xlarge instance is GPU-enabled and suitable for training, while the ml.c5.large instance is cost-optimized for CPU-based deployment scenarios.
- C. Incorrect.
Incorrect: Using a p3.2xlarge instance for both training and deployment will incur unnecessary costs during deployment, as GPUs are not typically required for inference.
- D. Incorrect.
Incorrect: The ml.m5.large instance is not suitable for training since it lacks GPU acceleration, and using a p3.2xlarge instance for deployment is costly and inefficient.
- E. Incorrect.
Incorrect: While the g4dn.xlarge instance is suitable for training, using it for deployment would result in higher costs compared to CPU-based alternatives like the ml.c5.large.