PostgreSQL Performance and Cost Analysis Across 23 AWS EC2 Instance Types

Executive Summary

For workloads requiring 33,000 requests per second (RPS) with a 1 GB dataset, the m8g.large instance using gp3-baseline storage is the most cost-effective configuration, costing approximately $82/month. While the c8i.large provides the highest raw throughput per dollar spent, Graviton-based instances (arm64) generally offer a superior balance of performance and cost for PostgreSQL deployments.

Cost-Performance Frontier for 33,000 RPS

When targeting a throughput of 33,000 RPS with a 1 GB dataset, 30 out of 52 tested configurations meet the requirement. The selection of the optimal instance depends on whether the priority is absolute minimum cost or maximum efficiency.

Cheapest Configuration

• Instance: m8g.large
• Storage: gp3-baseline
• Architecture: arm64 (Graviton)
• Cost: ~$82/mo
• Performance: 45,515 RPS (1.4× target)

Best RPS per Dollar

• Instance: c8i.large
• Storage: gp3-baseline
• Architecture: x86-64
• Performance: 52,203 RPS
• Efficiency: 451,476 RPS/€·hr

Comparative Analysis of Instance Families

Performance varies significantly across EC2 families, with newer generations and Graviton processors showing distinct advantages in throughput and latency.

Graviton (arm64) vs. x86-64

Graviton instances (highlighted in green in the source data) frequently appear on the cost-performance frontier. For example, the m8g.large (arm64) outperforms the m7i.large (x86-64) in both cost ($82/mo vs $90/mo) and throughput (45,515 RPS vs 41,305 RPS) when using gp3-baseline storage.

Performance Scaling by Instance Size

Increasing instance size generally improves throughput, though the RPS per dollar often diminishes as instances grow larger:

• m8g.large: 45,515 RPS (~$82/mo)
• m8g.xlarge: 70,871 RPS (~$147/mo)
• m8g.2xlarge: 81,451 RPS (~$278/mo)

Impact of Storage Configuration: gp3-baseline vs. gp3-max

Storage configuration significantly impacts the monthly cost without always providing a proportional increase in throughput.

Using the c8i.large as an example:

• gp3-baseline: 52,203 RPS at ~$84/mo
• gp3-max: 53,444 RPS at ~$184/mo

In this scenario, upgrading to gp3-max increased the monthly cost by over 100% while providing only a marginal increase in requests per second, suggesting that for this specific workload, baseline gp3 storage is sufficient.

Low-End Performance (T-Series Instances)

T-series instances (t3, t4g) are unsuitable for high-throughput PostgreSQL workloads. These instances consistently failed to meet the 33,000 RPS target, with most delivering between 2,800 and 5,300 RPS. For instance, the t4g.small on gp3-baseline only achieved 2,842 RPS, making it an order of magnitude slower than the m8g.large.