ClickHouse's PostgresBench pits Neon, Supabase, and Timescale against RDS

A new open-source benchmark from ClickHouse provides a reproducible test suite for managed Postgres providers. The initial results offer a performance snapshot, but the methodology warrants careful scrutiny.

The Answer Up Front

For founders and engineering leads evaluating managed Postgres providers, PostgresBench is a valuable new tool. Its real utility is not the specific leaderboard published by ClickHouse, but the open-source framework itself, which you can run against your own shortlist of vendors. Teams already committed to a provider with acceptable performance have no urgent need to act on these initial results. The bottom line: use the ClickHouse findings as a starting point to identify potential candidates like Neon, Supabase, or Timescale, but reserve final judgment until you benchmark them yourself using a configuration that mirrors your own application's workload. Never choose a database based on a single, third-party synthetic test.

Methodology

This v0 review analyzes PostgresBench, a new benchmarking tool, as of its announcement on June 20, 2026. The analysis is based exclusively on the introductory blog post published by ClickHouse, which details the tool's purpose, structure, and initial performance findings. The source signal is the post titled "PostgresBench: A Reproducible Benchmark for Postgres Services" available at https://clickhouse.com/blog/postgresbench.

This review covers the benchmark's stated methodology, the providers tested (Neon, Supabase, Timescale, AWS RDS), and the performance metrics reported by ClickHouse, specifically queries per second (QPS) and p99 latency across read-only, write-only, and mixed workloads. What is not covered is any independent verification of these results. We have not run PostgresBench ourselves. This analysis focuses on interpreting the vendor's claims and assessing the tool's utility for a technical founder. A future update will require independent, reproducible tests.

What It Does

An open-source framework

PostgresBench is an open-source suite of scripts and configurations built around pgbench, the standard benchmarking tool included with PostgreSQL. It aims to provide a standardized, reproducible way to measure the performance of any Postgres-compatible database service. The code is available on GitHub, allowing anyone to inspect the methodology and run the tests. The goal is to move beyond opaque vendor claims and enable direct, apples-to-apples comparisons.

Standardized workloads

The benchmark defines several distinct workloads to simulate different types of application behavior. The ClickHouse post focuses on three primary tests:

1. Read-Only: Simulates a caching layer or a high-traffic content delivery application.
2. Write-Only: Simulates an ingestion-heavy workload, like logging or event tracking.
3. Read-Write (80/20): A mixed workload representative of a typical transactional web application.

This multi-faceted approach provides a more nuanced performance picture than a single, generic test.

Initial provider comparison

The launch article presents results from running PostgresBench against four popular managed services: Neon, Supabase, Timescale, and Amazon RDS for PostgreSQL. ClickHouse documents the specific instance sizes and configurations used for each provider, which is critical for interpreting the results. For example, they compare serverless offerings with auto-scaling compute against traditional, provisioned instances. The published results include detailed charts comparing QPS and p99 latency for each provider across the different workloads.

What's Interesting / What's Not

Reproducibility is the point

The most important feature of PostgresBench is not the initial set of results, but its open and reproducible nature. By publishing the tool, ClickHouse is providing the community with a framework for verification. This puts pressure on database vendors to ground their performance marketing in verifiable data. For founders, this means you can (and should) move from trusting a vendor's homepage numbers to running a standardized test on your own.

A benchmark is not your workload

The primary caveat is that pgbench is a synthetic benchmark. Its default TPC-B-like workload does not represent the query complexity of most modern applications. The results published by ClickHouse are valid only for the specific, simple queries pgbench executes. An application with complex, multi-table joins or specific indexing strategies might yield entirely different performance rankings. The benchmark is a useful proxy for raw compute and I/O performance, but it is not a substitute for testing with your actual application.

Configuration is everything

Database performance is highly sensitive to configuration. The ClickHouse team reports their chosen configurations, but it's unclear if these represent the optimal setup for each provider. A seemingly small difference in connection pooling, memory allocation, or instance type can have an outsized impact on results. The initial findings should be viewed as a performance snapshot of one specific set of choices, not a definitive ranking of the platforms themselves.

Pricing

PostgresBench is an open-source tool available on GitHub and is free to use. Pricing for the tested database providers (Neon, Supabase, Timescale, AWS RDS) varies significantly based on usage, instance size, and features. (Pricing snapshot: June 20, 2026).

Verdict

PostgresBench is a welcome addition to the database tooling ecosystem. It provides a much-needed, open-source baseline for performance testing in a market crowded with competing claims. Founders should treat the initial results published by ClickHouse as a directional guide for which modern Postgres providers are worth evaluating. If you're choosing a new database, use these results to build a shortlist, then use the PostgresBench tool to run your own tests on configurations that match your expected production load. For those already on a platform like RDS, the benchmark offers a script to sanity-check if newer, serverless-first providers might offer a compelling performance-per-dollar advantage. Do not migrate based on these numbers alone; verify them for your specific workload.

What We'd Test Next

To move this from a v0 analysis to a v1 benchmark, we would need to run the suite ourselves. Our first step would be to reproduce the ClickHouse results on identical configurations to verify their findings. Next, we would test performance on different instance tiers to understand scaling characteristics and cost-performance trade-offs. We would also design a test to measure a crucial factor for serverless databases: cold start latency under intermittent load. Finally, we would customize the pgbench script with more complex queries (e.g., involving joins and text search) to see how the provider rankings shift when moving from a synthetic workload to one that more closely resembles a real-world SaaS application.

The investor read

PostgresBench signals a maturation of the DBaaS market. As core Postgres hosting becomes commoditized, vendors are competing on performance, developer experience, and workload-specific optimizations. This tool accelerates that trend by making performance claims more transparent and verifiable, forcing providers to compete on measurable results rather than marketing. The fact that ClickHouse, a competitor in the broader database market, is building tools for the Postgres ecosystem is a savvy strategic move. It positions them as an authority on data infrastructure and builds goodwill with developers. For investors, this means a company competing in the Postgres space needs a clear, benchmark-backed performance story for a target workload. A differentiated position on performance, cost-efficiency, or developer workflow for a specific niche is now table stakes.