MLA-C01 Question 298
Single answerYou are building a machine learning model on AWS SageMaker and need to choose the appropriate resource type for training. The model's training requirements are unpredictable, with varying compute needs based on the size of the training dataset and hyperparameter configurations. Which resource type is the most cost-effective and flexible for this scenario?
- A
On-demand instances
- B
Provisioned instances
- C
Spot instances
- D
Reserved instances
Show answer and explanation
Correct answer: A
Explanation
On-demand instances are the best choice for unpredictable workloads like machine learning training with varying compute requirements. They allow you to pay for compute resources without long-term commitment and provide the flexibility needed for dynamic workloads. Other options, such as spot and reserved instances, are either not suitable for unpredictable workloads or require upfront commitments that reduce flexibility.
- A. Correct.
On-demand instances are suitable for unpredictable workloads since they allow you to pay for compute capacity by the hour or second without any upfront commitment. This makes them cost-effective and flexible for workloads with varying requirements.
- B. Incorrect.
Provisioned instances are not a specific AWS compute offering. This term is incorrectly used in this context and does not describe an AWS resource type.
- C. Incorrect.
Spot instances can be cost-effective, but they are not ideal for unpredictable training needs where interruptions can disrupt progress. Spot instances are best for fault-tolerant and flexible workloads.
- D. Incorrect.
Reserved instances are designed for predictable workloads and require an upfront commitment, making them less flexible and cost-effective for unpredictable machine learning training requirements.