AWSOperational Efficiency and Optimization for GenAI Applications

Cost Optimization & Provisioned Throughput — AWS GenAI Dev Practice Question

A representative AWS Generative AI Developer (AWS GenAI Dev) exam question on Cost Optimization & Provisioned Throughput. Work through it below, then read why each option is right or wrong.

Short answer

The correct answer is D. Use on-demand pricing for all real-time requests, submit the Tuesday batch job using Amazon Bedrock batch inference at a 50% discount, and implement prompt optimization to reduce token counts.

Bedrock batch inference provides up to 50% cost reduction for non-time-sensitive workloads, making it ideal for the Tuesday summarization job (the largest cost driver at 50,000 requests). On-demand pricing is appropriate for the variable real-time traffic since provisioned throughput for peak would be wasted during off-peak hours. Prompt optimization (reducing unnecessary tokens) provides additional savings across all requests. This combination achieves the 30% cost reduction target.

The Question

A media company runs a content generation platform on Amazon Bedrock that has highly variable traffic patterns. During business hours (8 AM - 6 PM), the platform handles 5,000 requests per hour using Claude 3.5 Sonnet. During off-peak hours, traffic drops to 200 requests per hour. On Tuesdays, a batch job generates 50,000 article summaries. The company is currently using on-demand pricing and wants to reduce their monthly Bedrock costs by at least 30%. Which cost optimization strategy should the developer implement?

APurchase provisioned throughput with a 1-month commitment sized for peak business-hour traffic, and use cross-region inference profiles for the batch job to distribute load
BUse on-demand pricing for real-time traffic during business hours, switch to batch inference for the Tuesday summarization job, and use cross-region inference profiles for off-peak hours to access lower-cost regions
CPurchase provisioned throughput with a 6-month commitment sized for average traffic, and queue excess requests during peak hours using Amazon SQS with a Lambda consumer
DUse on-demand pricing for all real-time requests, submit the Tuesday batch job using Amazon Bedrock batch inference at a 50% discount, and implement prompt optimization to reduce token countsCorrect

Why D is correct

Bedrock batch inference provides up to 50% cost reduction for non-time-sensitive workloads, making it ideal for the Tuesday summarization job (the largest cost driver at 50,000 requests). On-demand pricing is appropriate for the variable real-time traffic since provisioned throughput for peak would be wasted during off-peak hours. Prompt optimization (reducing unnecessary tokens) provides additional savings across all requests. This combination achieves the 30% cost reduction target.

Why the other options are wrong

Option A: Purchase provisioned throughput with a 1-month commitment sized for peak business-hour traffic, and use cross-region inference profiles for the batch job to distribute load

Provisioned throughput sized for peak (5,000 req/hr) would be massively over-provisioned during off-peak hours (200 req/hr), wasting 96% of capacity for 14 hours per day. Cross-region inference profiles help with throughput limits but don't reduce costs — pricing is based on the region from which you call the profile.

Option B: Use on-demand pricing for real-time traffic during business hours, switch to batch inference for the Tuesday summarization job, and use cross-region inference profiles for off-peak hours to access lower-cost regions

Cross-region inference profiles don't provide access to lower-cost regions — the pricing is based on the calling region's rates, not the destination region. The rest of the strategy has merit but this fundamental misunderstanding makes it incorrect.

Option C: Purchase provisioned throughput with a 6-month commitment sized for average traffic, and queue excess requests during peak hours using Amazon SQS with a Lambda consumer

Provisioned throughput sized for average traffic would be insufficient during peak hours, causing degraded user experience. Queuing real-time user requests through SQS adds unacceptable latency for an interactive content generation platform. The 6-month commitment also reduces flexibility.

Key idea: Cost Optimization & Provisioned Throughput

Why D is correct: Bedrock batch inference provides up to 50% cost reduction for non-time-sensitive workloads, making it ideal for the Tuesday summarization job (the largest cost driver at 50,000 requests). On-demand pricing is appropriate for the variable real-time traffic since provisioned throughput for peak would be wasted during off-peak hours. Prompt optimization (reducing unnecessary tokens) provides additional savings across all requests. This combination achieves the 30% cost reduction target. Why A is wrong: Provisioned throughput sized for peak (5,000 req/hr) would be massively over-provisioned during off-peak hours (200 req/hr), wasting 96% of capacity for 14 hours per day. Cross-region inference profiles help with throughput limits but don't reduce costs — pricing is based on the region from which you call the profile. Why B is wrong: Cross-region inference profiles don't provide access to lower-cost regions — the pricing is based on the calling region's rates, not the destination region. The rest of the strategy has merit but this fundamental misunderstanding makes it incorrect. Why C is wrong: Provisioned throughput sized for average traffic would be insufficient during peak hours, causing degraded user experience. Queuing real-time user requests through SQS adds unacceptable latency for an interactive content generation platform. The 6-month commitment also reduces flexibility. On the AWS GenAI Dev exam, questions in the "Operational Efficiency and Optimization for GenAI Applications" domain test whether you can map a scenario's constraints to the right choice. Read the requirement carefully, eliminate options that violate any single constraint, and pick the one that satisfies all of them with the least operational overhead.

Ready to see how you'd score?

Take the free practice quiz and find out which AWS Generative AI Developer - Professional domains you need to focus on. No signup required.

Practice 5 similar questions

Same cert, same or adjacent domain. Use these after reviewing the explanation.

Related AWS Generative AI Developer - Professional Practice Questions