home/product/performance
[ performance ] built for speed

Blazing-fast NVMe.
A fraction of the cost.

Fast Postgres isn't magic — it's the right hardware with sensible defaults. Local NVMe instead of network disk, dedicated CPU and RAM on Pro and up, pooling built in, and PostgreSQL 17 tuned per instance size. Low p99 latency, none of the big-cloud bill.

[01] why it's fast

Speed that's structural, not lucky.

Every layer is chosen for low latency under real load — from the disk your data sits on to the connection pooler in front of it.

Local NVMe storage

Your data lives on local NVMe attached to the box — not slow network-attached disk. High IOPS and low latency are the baseline, not a paid add-on.

Dedicated CPU & RAM

On Pro and up, cores and memory are reserved for your database alone — no noisy neighbours stealing cycles at the worst possible moment.

Tuned for Postgres 17

The latest PostgreSQL with sensible defaults per instance size — shared_buffers, work_mem and autovacuum set for your hardware, so you start fast and stay fast.

Connection pooling built in

PgBouncer ships with every database, so serverless and high-concurrency apps don't exhaust your connections. No separate service to run or patch.

Read replicas

Spread read traffic across replicas and serve data closer to your users. Spin one up from the dashboard the moment reads become your bottleneck.

Online upgrades

Resize CPU, memory and disk and run major-version upgrades online — handled with you, without the downtime drama or the maintenance-window dread.

[02] latency

Lower p99, the day you move.

What teams typically see when they leave a busy, shared instance for dedicated NVMe: the spikes flatten and the tail latency drops.

// p99 query latency — moving from RDS to PostgresRun
RDS PostgresRun 3000ms 2000ms 1000ms 0 migrated on RDS on PostgresRun 12:00 12:30 13:00 13:30 14:00 14:30
p99 latency drops sharply when you move from a busy, shared RDS instance to dedicated NVMe. Illustrativeask us for real before/after numbers from your own migration.
[03] by the numbers

The shape of fast.

The headline figures that follow from the hardware and tuning above. Numbers marked with a tilde are illustrative.

Local NVMe
High-IOPS storage, standard on every plan
~50%
Lower cost than AWS RDS for the same managed Postgres
<2 min
From zero to a ready, production database
P17
Tuned PostgreSQL 17, sensible defaults per size
[04] methodology

How we talk about numbers.

We won't pretend a single benchmark settles anything. Database performance depends on your schema, your query mix, your indexes and your working-set size — so any chart or figure on this page is illustrative, meant to show the shape of what changes when you move from a busy, shared instance to dedicated local NVMe, not a promise of an exact millisecond count.

What's a hard fact: every database runs PostgreSQL 17 on local NVMe, Pro and up get dedicated CPU and RAM, PgBouncer is built in, and instances are tuned for their size. What's illustrative: the latency curve, the percentages, and the "fraction of the cost" framing — directionally true, not laboratory-precise.

Want numbers you can actually trust for your workload? We'll run a test migration of a representative database and show you real before/after p99 latency and cost. Tell us what you're running and we'll set it up.

[05] questions

Performance FAQ.

Is local NVMe durable?
Yes. Speed doesn't come at the cost of safety: every database has off-site, encrypted backups with point-in-time recovery, and on higher plans you can add a synchronous HA standby with automatic failover. Local NVMe gives you the IOPS; backups and replication give you the durability.
What about noisy neighbours?
On Pro and up, CPU and memory are dedicated to your database alone — nobody else's workload can steal your cycles. Smaller plans use burstable CPU, which is great for dev and light production; if you need guaranteed performance under load, dedicated CPU starts at Pro.
Do upgrades cause downtime?
Major-version upgrades and resizes are online and assisted — we plan them with you and run them with minimal disruption, not a scary maintenance window. No copy-paste runbooks, no downtime drama.
Can I see real numbers for my workload?
Yes — that's the honest answer. The charts here are illustrative, so we'll run a test migration of a representative database and show you actual before/after p99 latency and cost. Get in touch and we'll set it up.
provision in 120 seconds

Feel the difference.

Spin up a database on dedicated NVMe in two minutes — or let us migrate a test workload and show you the real before/after.