4.1.3 · Lesson 3 of 16
Throughput Optimization & Batch Inference
Maximize processing capacity for non-real-time workloads using Bedrock Batch Inference and discounted async execution.
WHAT
Maximize system processing capacity for workloads that do not need real-time responses.
HOW
- Use Amazon Bedrock Batch Inference for asynchronous jobs (e.g., overnight summary processing).
- Expect roughly a 50% discount versus standard on-demand API calls.
EXAM TRAP
Do not choose real-time Converse for massive overnight batch jobs when cost is the primary constraint.
Key points
- Bedrock Batch Inference
- 50% discount vs on-demand
- Overnight / async jobs
- S3 input manifests
- Non-real-time summarization
Exam focus
For non-real-time ingestion (e.g., nightly summaries), Bedrock Batch Inference is the high-yield low-cost path — about 50% off on-demand.