SAP-C03 cost-optimized design explained
Cost-Optimized Architecture Design is the third of the five SAP-C03 domains and carries 15% of scored content. It asks an architect to meet requirements at the lowest sensible cost without breaking resilience, security or performance targets, which means knowing what drives each bill and which lever changes it. The levers below are aligned with the Cost Optimization pillar of the AWS Well-Architected Framework and the four tasks in the SAP-C03 guide.
What the domain asks of an architect
- Purchase model. On-Demand has no commitment. Savings Plans commit to hourly spend: Compute Savings Plans apply across instance families, Regions, and Fargate and Lambda, while EC2 Instance Savings Plans fix the family in a Region for a deeper discount. Reserved Instances (Standard or Convertible) commit to specific capacity. Spot Instances are cheapest but can be interrupted, so they suit fault-tolerant, stateless or batch work.
- Right-sizing. AWS Compute Optimizer recommends sizes from usage. Graviton-based instances often improve price-performance. Serverless removes idle cost for spiky workloads but can cost more for steady, high-volume ones.
- Storage. S3 storage classes and lifecycle rules, S3 Intelligent-Tiering for unknown access patterns, EBS volume types, deleting unattached volumes and old snapshots.
- Data transfer. Traffic between Availability Zones, across Regions, out to the internet and through NAT gateways is charged. Gateway VPC endpoints for S3 and DynamoDB carry no hourly or processing charge and keep that traffic off NAT gateways.
- Visibility. Cost allocation tags, AWS Cost Explorer, AWS Budgets, Cost Anomaly Detection and consolidated billing in Organizations make cost attributable.
The four tasks in the AWS exam guide
| Task | What it covers |
|---|---|
| 3.1 Design architectures based on AWS pricing models | Purchase options (provisioned capacity, Reserved Instances, Savings Plans, Spot, On-Demand), Graviton and Step Functions Standard vs Express, rightsizing with Compute Optimizer, scheduled and target-tracking scaling, Fargate Spot, AI/ML cost levers (service tier, batching, caching) |
| 3.2 Optimize data transfer and network costs | CloudFront and Origin Shield, private connectivity and VPC endpoints, NAT gateway and inter-AZ and inter-Region traffic, cost-aware replication, Transit Gateway and PrivateLink pricing, cross-Region inference routing for AI/ML |
| 3.3 Design cost allocation and visibility | Tagging strategies, Cost Explorer and Cost and Usage Reports, tag enforcement with SCPs, Config rules and tag policies, chargeback and showback (cost categories, split cost allocation, Bedrock token usage), Budgets and Cost Anomaly Detection |
| 3.4 Optimize storage and database costs | S3 Intelligent-Tiering, lifecycle and Glacier classes, EFS lifecycle, EBS gp3 and Provisioned IOPS, DynamoDB capacity modes, reserved DB instances, Aurora Serverless and I/O-Optimized, Database Savings Plans, AWS Backup and Data Lifecycle Manager retention, cost-effective observability |
New compared with SAP-C02, per the guide: AI/ML cost optimization (tasks 3.1 and 3.2) and cost-effective observability (log filtering, metric aggregation, retention) in task 3.4. Several SAP-C02 cost skills, spread over its other domains, were consolidated into this domain. The pricing details to check are always on the AWS pricing pages.
Worked mini-scenario
A media company processes uploaded video. A baseline of 20 encoder instances runs all day; peaks add up to 80 more for two hours. Output is archived in S3 and read rarely after 30 days.
- Baseline: cover the steady 20 with a Compute Savings Plan, which stays valid if the family changes.
- Peak: run the extra capacity on Spot Instances across several instance types, with the job queue handling interruptions by retrying. Reserved capacity for a two-hour peak would sit idle most of the day.
- Storage: lifecycle rules move outputs to an infrequent-access or archive class after 30 days. If retrieval time matters, an archive class with minutes-to-hours retrieval may be unacceptable, so confirm the requirement.
Trade-off: the savings plan commits spend even if the baseline shrinks, and Spot adds retry complexity. The saving is only valid if the workload tolerates it.
Easy-to-confuse pairs
| Pair | The distinction |
|---|---|
| Savings Plans vs Reserved Instances | Spend commitment with flexibility vs capacity-specific commitment |
| Compute vs EC2 Instance Savings Plans | Flexible across family and Region vs fixed family in a Region |
| Standard vs Convertible Reserved Instances | Higher discount, fixed attributes vs exchangeable attributes |
| S3 Intelligent-Tiering vs lifecycle rules | Automatic by observed access vs scheduled by age |
| Gateway vs interface VPC endpoint | Free route-table endpoint for S3 and DynamoDB vs hourly and per-GB PrivateLink |
| Budgets vs Cost Explorer vs Anomaly Detection | Alert on thresholds vs analyse history vs detect unusual spend |
Exam-style pitfall
Optimising the cheapest line while violating a stated requirement. Spot for a stateful database, archive classes for data read daily, or a three-year commitment for a workload that is about to be re-platformed all look cheap but fail the scenario. The correct answer is the cheapest option that still satisfies every requirement.
Practice questions
Question 1. Workloads in private subnets read several terabytes a day from Amazon S3 in the same Region through a NAT gateway, and the NAT data processing charge is large. Which change reduces cost the most without changing the application?
- A. Replace the NAT gateway with a larger one
- B. Add a gateway VPC endpoint for Amazon S3 and update the route tables
- C. Add a second NAT gateway in another Availability Zone
- D. Copy the data to another Region
Show answer
Answer: B
A gateway VPC endpoint for S3 routes the traffic privately through the route table at no hourly or processing charge, so it no longer passes through the NAT gateway. A larger NAT gateway does not remove the per-GB charge, an extra NAT gateway adds cost, and moving the data to another Region adds transfer charges.
Want more questions like this? Full SAP-C03 practice tests →
Question 2. A batch pipeline renders images from a queue. Jobs are idempotent and can be retried, and the finish time is flexible. The team wants the lowest compute cost. Which option is best?
- A. On-Demand Instances in an Auto Scaling group
- B. Three-year Standard Reserved Instances
- C. A fleet of Spot Instances across several instance types
- D. Dedicated Hosts
Show answer
Answer: C
Spot Instances give the deepest discount and suit interruption-tolerant work, and an idempotent job queue lets interrupted jobs be retried. On-Demand costs more, a three-year Reserved Instance commits capacity that a flexible batch job does not need, and a dedicated host costs the most.
Want more questions like this? Full SAP-C03 practice tests →
Question 3. Finance cannot tell which team causes each part of a shared account's bill. The company wants costs grouped by team and project with minimal change to workloads. What should the architect do?
- A. Define and activate cost allocation tags and enforce them with tag policies
- B. Create one AWS Budget for the whole account
- C. Resize all instances to smaller types
- D. Buy Reserved Instances for all running instances
Show answer
Answer: A
Applying and activating cost allocation tags, enforced with tag policies, lets Cost Explorer and the cost and usage report group spend by team and project. A budget alone alerts on totals, resizing instances does not attribute cost, and Reserved Instances change price rather than visibility.
Want more questions like this? Full SAP-C03 practice tests →