ClickHouse built PostgresBench to fill a gap in reproducible database benchmarking. The company, known for its analytical database performance, applied the same transparent methodology from its Click Bench project to transactional Postgres workloads. This addresses the lack of standardized comparisons for managed Postgres services.

Postgres Bench uses pgbench with a TPC-B-like workload simulating concurrent transactions. Tests run with 256 clients and 16 threads for 10 minutes across two scale factors: 6,849 records (~100 GB) and 34,247 records (~500 GB). These sizes represent different growth stages where the smaller fits in cache while the larger forces disk I/O, revealing how services handle storage pressure.

Five managed Postgres services were tested in us-east-2 with HA disabled: ClickHouse's own offering, AWS Aurora PostgreSQL, AWS RDS for PostgreSQL, Neon, and Crunchy Bridge. All run on Graviton instances with NVMe caching where available. Each service was tested at small (4 vCPU/16GB) and large (16 vCPU/64GB) configurations.

ClickHouse's managed Postgres led the results with 28,668 TPS on large instances versus 12,628 TPS for Aurora. The benchmark publishes all queries, datasets, configurations, and raw results openly. Anyone can validate findings or submit corrections, making this a transparent alternative to vendor-provided benchmarks that often lack reproducibility.