Price-Performance Evaluation of AWS Graviton vs. Intel-based Aurora PostgreSQL Instances
Abstract
AWS Graviton processors present cloud architects with a critical cost-performance decision for relational databases, yet independent validation for real-world Online Transaction Processing (OLTP) workloads on managed services like Amazon Aurora PostgreSQL remains limited. This study provides a tier-aware evaluation comparing Graviton and Intel architectures across Aurora’s general-purpose (M-series) and burstable (T-series) instance families. We conduct controlled benchmarks using pgbench on architecturally equivalent general-purpose instances (db.m7g.large vs. db.m7i.large) and relevant burstable instances (db.t4g vs. db.t3). Results, validated with high statistical consistency (Coefficient of Variation ¡ 3%), reveal a decisive tier-dependent outcome. For general-purpose computing, Intel’s db.m7i.large delivers 29–68% higher throughput depending on workload and concurrency level and a 22.6% better cost-per-transaction ratio than the Graviton3-based db.m7g.large, despite a 6% higher hourly cost. Conversely, in the burstable tier, the newer Graviton2-based T4g instances provide 4–30% higher throughput, depending on instance size, and 18–29% better cost-performance compared to the older Intel T3 instances they commonly replace. This burstable advantage reflects both architectural improvements and a generational leap. Our findings demonstrate that optimal architectural selection depends on both instance family and generation, providing cloud architects a data-driven framework: choose Intel for performance-critical general-purpose workloads, and migrate to Graviton for cost-optimized burstable deployments replacing previous-generation Intel instances.