A company is building custom deep learning models in Amazon SageMaker by using training and inference containers that run on Amazon EC2 instances. The company wants to reduce training costs but does n

Question

Options

• AOn-Demand Instances
• BCheckpoints
• CReserved Instances
• DIncremental training
• ESpot instances

• A
  6% (1)
• B
  72% (13)
• C
  17% (3)
• D
  6% (1)

Explanation

Spot Instances (E) offer up to 90% cost savings over On-Demand pricing by using unused EC2 capacity, making them the most cost-effective compute option. The key constraint - that training jobs can finish after interruptions - makes Spot Instances viable here, since they can be reclaimed by AWS at any time. Checkpoints (B) are the essential complement: they periodically save training state to S3, so when a Spot Instance is interrupted, the job can resume from the last checkpoint rather than restarting from scratch. Together, these two features enable fault-tolerant, low-cost training with no architectural changes. On-Demand Instances (A) are more expensive. Reserved Instances (C) require a 1–3 year commitment and don't reduce per-job cost for variable workloads. Incremental training (D) reuses a previously trained model's artifacts to continue training - it's a technique for reducing training time/data, not compute cost.

No community discussion yet for this question.