2026 Postgres Benchmarks: RDS vs Hetzner vs Managed

Two benchmarks, two different stories about Postgres hosting costs

If you're running Postgres in production, two 2026 benchmarks are worth a look before you renew your database contract or scale up your instance. One, from Hostim.dev, ran in July 2026 on PostgreSQL 16 in a central-Europe region and directly compared AWS RDS, self-hosted Postgres on Hetzner, and Hostim's own managed offering. The other, from postgres.saneengineer.com, tested 23 EC2 instance types running PostgreSQL 17 in us-east-1 under on-demand pricing — but doesn't mention Hetzner or any self-hosted comparison at all.

Sources differ on PostgreSQL version (16 vs. 17), region (central-Europe vs. us-east-1), and even scope — it's unclear whether these are related studies or entirely independent efforts. Treat the two data sets separately.

The Hostim.dev numbers

At matching specs (2 vCPU/4GB, PostgreSQL 16), Hostim.dev reported:

2.5x the write throughput of AWS RDS db.t4g.medium
2.1x the write throughput of default self-hosted Postgres on Hetzner

The test setup used a pgbench dataset at scale factor 50 (~750 MB), a write-heavy run with 4 clients over 300 seconds, a read-only run with 8 clients over 300 seconds, and a single-connection write latency test over 60 seconds.

Important caveat: these are relative multipliers only — no absolute TPS or latency-in-milliseconds figures are provided, so it's hard to know how any of the three options would perform under your specific query patterns or connection concurrency.

What it costs

The cost figures tell a more concrete story:

AWS RDS db.t4g.medium (instance only, single-AZ, on-demand): ~$48/month
AWS RDS Provisioned IOPS SSD (3000 IOPS default): ~$300/month
Hostim's drp-50 replicated tier (standby + automatic failover included): €50/month
Self-hosted Hetzner, second VPS for replication: €20/month — but this doesn't include the engineering time to configure failover

The report flags that RDS's real-world cost can run to a multiple of the listed instance price once storage, IOPS, egress, and Multi-AZ charges are added — the jump from ~$48 to ~$300/month for provisioned IOPS illustrates that clearly.

The other benchmark: EC2 instance types

Separately, postgres.saneengineer.com benchmarked 23 EC2 instance types under PostgreSQL 17, using a 90% read / 10% write workload at concurrency level 32. The report doesn't specify which instance types came out on top or bottom, and no cost figures are included for this benchmark — so it's most useful as a signal that instance-type choice matters, not as a ready-made buying guide.

The risks to keep in mind

RDS's burstable db.t4g.medium instance relies on a CPU-credit model; sustained workloads that exhaust those credits may see performance throttle.
Self-hosted Hetzner setups with replication require manually building and operating failover tooling like Patroni or repmgr — a real operational burden, not just a line-item cost.
Listed RDS instance prices routinely understate total spend once storage, IOPS, egress, and Multi-AZ are factored in.

Why founders should care

For early-stage teams choosing or re-evaluating a Postgres host, this data suggests a few probabilistic takeaways rather than firm conclusions:

Teams running write-heavy workloads on small RDS instances may plausibly find better price-performance with managed alternatives, though the lack of absolute throughput numbers means this shouldn't be treated as certain.
Founders budgeting for RDS should likely plan for costs well beyond the base instance price — the ~$48 to ~$300/month jump for IOPS alone suggests total spend could be several times the sticker price.
Teams considering self-hosted Hetzner to save on the dollar cost of replication should factor in the engineering hours needed to build and maintain failover tooling, which isn't captured in the raw price comparison.
Given the differing PostgreSQL versions and regions between the two benchmarks, founders should be cautious about assuming either result set will generalize to their own production environment, workload shape, or region.

Where to go from here

Hostim offers a free database tier and a 5-day trial with no card required, which could let founders test fit at low risk before committing. That said, the report doesn't include absolute performance numbers, variance/error margins, or repeated-trial data for either benchmark — so any evaluation should include your own load testing against your actual query patterns before making a hosting decision.