AMD EPYC 7543P32 cores · 256 GB DDR4 ECC · 2 × 1.92 TB NVMe
The classic database box: 32 cores, 256 GB and 256 MB of L3 cache for busy OLTP.
Databases
Databases deserve the whole machine
PostgreSQL, MySQL, ClickHouse and Redis reward three things: memory, fast and consistent storage, and cores nobody else uses. Bare metal gives you all three — with ECC so a flipped bit never reaches your data.
- Up to 1 TBECC memory
- NVMeno IOPS cap, no noisy neighbour
- ECCon every server
- $249from, per month
What this workload needsWorkload profile
- Memory, and lots of itThe working set should live in RAM. Up to 1 TB on our dual-EPYC server keeps even large datasets hot.
- Predictable latencyShared cloud volumes vary with other tenants’ load. Local NVMe gives consistent sub-millisecond latency for WAL and random reads.
- Cores for parallel queriesAnalytical engines like ClickHouse parallelise across every core you give them; OLTP benefits from high clocks.
- ECC, alwaysEvery server we sell has ECC memory. Silent corruption is the one failure a database cannot detect on its own.
Recommended
The servers we would deploy for it
Chosen by the engineers who size these workloads every week. Every one of them is in stock.
AMD EPYC 9455P48 cores · 128 GB DDR5 ECC · 2 × 1.92 TB NVMe
DDR5 across 12 channels and 48 Zen 5 cores for analytics; upgrade to 768 GB on order.
2 × AMD EPYC 753264 cores · 1 TB DDR4 ECC · 2 × 1.92 TB NVMe
In-memory at scale: 1 TB of RAM across two sockets for Redis, caches and huge buffer pools.
Setting it up
Advice from our engineers
The first hour on a new server decides how the next year goes.
Where to host it
Keep the database next to its application servers. For a replica, choose a second city in the same region — Amsterdam and Frankfurt are a few milliseconds apart.
Compare latency from your users- Start with shared_buffers at ~25% of RAM for PostgreSQL and let the OS cache the rest.
- Enable huge pages for large buffer pools and pin the database to NUMA nodes on dual-socket servers.
- Keep WAL/binlog on NVMe and archive it off the server continuously (pgBackRest, WAL-G).
- Replicate across locations: a streaming replica in another city turns a disaster into a failover.
FAQ
Questions about databases
Still have a question?
Ask the engineers who run the servers — median first reply under 15 minutes, 24/7.
Contact usCan I get more memory than listed?
Yes. Most servers offer memory upgrades on the order page, installed in place later if you need them.
Do you offer managed databases?
No — you keep full control of the engine and its configuration. We keep the hardware, network and spare parts in shape.
Hardware RAID or software RAID?
We use Linux software RAID (mdadm) on NVMe: it is as fast as a controller for NVMe, portable, and you can monitor it directly.
Other workloads
All workloads