ClickHouse released PostgresBench on April 2, 2026, an open-source tool designed to provide transparent, reproducible performance metrics for managed PostgreSQL services. The benchmark utilizes pgbench, the standard tool for simulating short concurrent transactions, to measure transactional throughput and latency under consistent conditions. Each test run lasts 600 seconds with 256 clients and 16 threads, capturing metrics including average TPS, average latency, P95 latency, and P99 latency. The initial cohort of tested services includes Postgres managed by ClickHouse, AWS Aurora, AWS RDS, Neon, and Crunchy Bridge.

Testing encompassed two dataset scales, 6849 (~100 GB) and 34247 (~500 GB), to evaluate how different storage architectures handle data growth. For the 16 vCPU / 64 GB configurations at the 100 GB scale, ClickHouse's managed Postgres service achieved 28668 TPS with 8.908 ms average latency, compared to 12628 TPS (20.242 ms) for AWS Aurora, 8133 TPS (31.435 ms) for AWS RDS, 8563 TPS (29.832 ms) for Neon, and 14790 TPS (17.269 ms) for Crunchy Bridge. At the 500 GB scale, ClickHouse's service recorded 26328 TPS with 9.703 ms average latency, outpacing competitors including AWS Aurora (10402 TPS) and Crunchy Bridge (11113 TPS).

The benchmark design emphasizes fairness by using consistent hardware ratios and running all tests in the us-east-2 region with HA disabled to isolate compute and storage performance. Results are published in structured JSON, allowing users to verify methodologies or submit their own configurations via pull requests. ClickHouse attributes its performance lead to the use of NVMe storage co-located with compute, which avoids the network latency inherent in shared storage architectures like EBS or object storage. The tool, available on GitHub, aims to serve as a standardized reference for transactional database performance.

ClickHouse released PostgresBench on April 2, 2026, an open-source tool designed to provide transparent, reproducible performance metrics for managed PostgreSQL services. The benchmark utilizes pgbench, the standard tool for simulating short concurrent transactions, to measure transactional throughput and latency under consistent conditions. Each test run lasts 600 seconds with 256 clients and 16 threads, capturing metrics including average TPS, average latency, P95 latency, and P99 latency. The initial cohort of tested services includes Postgres managed by ClickHouse, AWS Aurora, AWS RDS, Neon, and Crunchy Bridge.

Testing encompassed two dataset scales, 6849 (~100 GB) and 34247 (~500 GB), to evaluate how different storage architectures handle data growth. For the 16 vCPU / 64 GB configurations at the 100 GB scale, ClickHouse's managed Postgres service achieved 28668 TPS with 8.908 ms average latency, compared to 12628 TPS (20.242 ms) for AWS Aurora, 8133 TPS (31.435 ms) for AWS RDS, 8563 TPS (29.832 ms) for Neon, and 14790 TPS (17.269 ms) for Crunchy Bridge. At the 500 GB scale, ClickHouse's service recorded 26328 TPS with 9.703 ms average latency, outpacing competitors including AWS Aurora (10402 TPS) and Crunchy Bridge (11113 TPS).

The benchmark design emphasizes fairness by using consistent hardware ratios and running all tests in the us-east-2 region with HA disabled to isolate compute and storage performance. Results are published in structured JSON, allowing users to verify methodologies or submit their own configurations via pull requests. ClickHouse attributes its performance lead to the use of NVMe storage co-located with compute, which avoids the network latency inherent in shared storage architectures like EBS or object storage. The tool, available on GitHub, aims to serve as a standardized reference for transactional database performance.