Q70 — AWS AIF-C01 Ch.2

Question 70 of 100 | ← Chapter 2

A company collects large tabular datasets weekly for its ML workflow. The workflow can tolerate up to 7 days of latency between data ingestion and inference. Inference may take several hours. The company wants a solution that performs inference on these datasets without incurring endpoint costs when idle. Which type of inference satisfies these requirements?

• A. Real-time Inference
• B. Batch Transform ✓
• C. Serverless Inference
• D. Asynchronous Inference

Correct Answer: B. Batch Transform

Explanation

B. Batch Transform is the correct choice. A. Real-time inference requires low-latency responses (milliseconds to seconds) and continuous endpoints—contradicting the 7-day latency tolerance and cost-efficiency requirement. B. Batch Transform processes large volumes of data offline, scales automatically, incurs no cost when idle, and supports long-running jobs—perfect for weekly, high-volume, latency-tolerant inference. C. Serverless inference (e.g., SageMaker Serverless Inference) auto-scales for sporadic requests but charges per invocation and duration—less cost-effective for large, infrequent batches. D. Asynchronous inference (e.g., SageMaker Asynchronous Inference) decouples request submission from execution but still requires provisioned endpoints and incurs idle costs—unlike Batch Transform, which uses ephemeral compute only during job execution. Thus, Batch Transform best matches the requirements: periodic large-scale inference, high latency tolerance, and zero idle cost.