A company runs training jobs on Amazon SageMaker by using a compute optimized instance. Demand for training runs will remain constant for the next 55 weeks. The instance needs to run for 35 hours…

Question

Options

• AUse a serverless endpoint with a provisioned concurrency of 35 hours for each week. Run the
• BUse SageMaker Edge Manager for the training. Specify the instance requirement in the edge
• CUse the heterogeneous cluster feature of SageMaker Training. Configure the instance_type,
• DOpt in to a SageMaker Savings Plan with a 1-year term and an All Upfront payment. Run a

• A
  4% (1)
• B
  7% (2)
• C
  15% (4)
• D
  74% (20)

Explanation

Option D is correct because SageMaker Savings Plans are specifically designed for predictable, steady workloads - exactly this scenario. By committing to a 1-year term with All Upfront payment, the company receives the maximum discount (up to ~64% off on-demand rates). The usage pattern (35 hrs/week × 55 weeks) is both consistent and long-lived enough to make the commitment financially worthwhile.

Why the distractors are wrong:

• A - Serverless endpoints with provisioned concurrency are for inference, not training. Training jobs run on dedicated compute instances; there's no serverless training option in SageMaker.
• B - SageMaker Edge Manager manages and deploys models on edge devices (IoT, embedded hardware). It has nothing to do with cloud-based training job cost optimization.
• C - Heterogeneous clusters let you mix instance types (e.g., CPU + GPU nodes) within a single training job for performance reasons, but they don't offer any pricing discount mechanism.

Memory tip: On AWS exams, whenever you see "constant/predictable demand over months or years," the answer almost always involves a commitment-based pricing model (Savings Plans, Reserved Instances). The longer the commitment and the more upfront you pay, the bigger the discount - match that pattern to eliminate the other distractors.

No community discussion yet for this question.