Linux turns 35 this week. Celebrate with 10% off for life. code LINUX35. Claim 10% Linux is 35: 10% for life · LINUX35 Claim 10%
Skip to content
Sign up
VPS Infrastructure

Infrastructure
that does what it says.

Run your servers as you run your local development environments: Standard, servers with dedicated cores and servers with a memory-hungry load as separate tiers, all on KVM, all on NVMe RAID10 storage, and published benchmarks, in Amsterdam and New York.

First month $1. A paid first month, not a free trial: after that the normal monthly price, one per customer.

Storage
NVMe RAID10
Storage IOPS
122k 4K random read
Regions
2 AMS, NYC
Start for $1 See plans Read benchmarks
Three tiers, picked on purpose

Three CPU tiers. Real differences.

This is NOT a case of artificial sub-tiers created for a 'Pro' or 'Scale Up' version of existing gear, to support specific workloads. It's three different hardware generations (3 tiers of CPU) created to support 3 different types of workloads. Scale up within a tier as needed, rather than scaling up between tiers.

Standard

S-2...S-64

Shared vCPU on EPYC Milan platform with DDR4 ECC memory. Suitable for web applications, SaaS platforms, development / staging environments.

  • CPUEPYC Milan (shared vCPU)
  • RAMDDR4 ECC
  • StorageNVMe RAID10
  • Ratio1:2, 1:4 vCPU
  • Plans4 GB to 64 GB of RAM
Best for: web apps, SaaS, dev/staging
From $6.79/mo (2yr discount)See Standard plans

Dedicated CPU

D-4...D-64

Pinned cores on EPYC Genoa processors with high-performance DDR5 ECC memory and NVMe RAID10 storage for latency-sensitive workloads such as databases and APIs.

  • CPUEPYC Genoa, pinned cores
  • RAMDDR5 ECC
  • StorageNVMe RAID10
  • SharingNone, cores reserved
  • Plans4 GB / 8 GB / 16 GB / 32 GB / 64 GB of RAM
Best for: databases, APIs and other CPU-bound applications
From $29.60/moSee Dedicated plans

High Memory

H-64...H-192

EPYC Milan + DDR4 ECC for Memory Intensive workloads. 1:8 RAM:vCPU ratio for In-Memory applications.

  • CPUEPYC Milan, shared vCPU
  • Memorylarge pool of DDR4 ECC
  • StorageNVMe RAID10
  • Ratio1:8 RAM:vCPU
  • Plans64 GB, 128 GB, 192 GB
Best for: databases, caching, analytics
From $199.17/moSee High Memory
Stack & numbers

Every component. Every measurement.

We publish every component and every measurement from our public benchmark results page.

CPU

Silicon
  • SEPYC Milan, shared vCPU, 1:2 to 1:4 ratio
  • DEPYC Genoa, pinned cores, no sharing
  • HMEPYC Milan, shared vCPU, 1:8 RAM ratio
Benchmark performance numbers (Geekbench 6 median) compared to OVH, Vultr and DigitalOcean.

Memory

ECC
  • SDDR4 ECC (hard-capped per node)
  • D4800 MT/s DDR5 ECC, paired with the Genoa core complex
  • HMDDR4 ECC memory, large pool, 1:8 ratio

Errors on for all memory. However the amount of memory per node does not scale with the amount of nodes you create (this is different from most hosting providers).

Storage

NVMe

RAID10, PCIe direct

Local arrays on each node (in fact local arrays on each core). No SATA here. No SANs. No sharing. Not even when it's loaded heavily, the latency is always sub-millisecond.

122k 4K random read IOPS @ 480 MiB/s, fio, D-tier 2 vCPU / 4 GB.

Network

Uplink

Multi-homed, IPv4 + IPv6

Connectivity to both data centers via multiple redundant uplink connections on all layers (VLAN's etc) via firewalls.

  • ALLDual-stack v4 + v6 on every plan

Virtualization

Hypervisor

KVM, separate kernel per VPS

Full hardware virtualization, with full kernel isolation. We don't offer virtualization in a 'cluster' configuration, like you'd find with OpenVZ or LXC for example. We do support nested virtualization.

  • ALLSame hypervisor across all tiers and regions

Regions

Locations

Amsterdam, New York

2 regions: Amsterdam (EU) and New York (US-East). Same Stack, same Control Plane. Same Pricing in EU and US-East.

  • EUAmsterdam, 52.37° N
  • USNew York, 40.71° N

Database & HTTP, measured

Top of compared

Here the medians of the measured metrics of pgbench TPS and wrk requests/second for the Database & HTTP workloads on 2 vCPU / 4 GB / Dedicated hosts in Amsterdam and New York. All runs can be found here.

See all runs
Two regions, one stack

Same network design. Same hardware. Different continents.

Multi-homed transit routers exist in both regions. Serve your users from the region that is closest to them or failover to both regions.

Amsterdam

NL, EU

52.37° N, 4.90° E

  • Dense peering and transit environment
  • Strong peering with networks in NL, BE, DE, UK and FR
  • EU-first SaaS / API workloads
VPS Amsterdam details

New York

US, East

40.71° N, 74.01° W

  • US East-Coast audience proximity
  • Multi-homed upstream, redundant transit options
  • SaaS, APIs and transactional sites
VPS New York details
Uplinks
Redundant, multi-homed
IP
IPv4 + IPv6 dual-stack
VLAN
Available on request
Firewall
Edge filtering, abuse policy
Operating systems

Linux-first. ISO upload supported.

This environment supports a number of distros and you can upload the ISO for the desired distribution to install the server in seconds. Then you can switch to another distro in a couple of seconds in the control panel or via API. You just need to reinstall the server in the chosen distribution.

Ubuntu

LTS & current

Default OS for web applications and SaaS workloads is Ubuntu.

Debian

Stable

Production reliability and updates that are very predictable.

AlmaLinux

RHEL-compat

Enterprise drop-in for RHEL workloads.

Rocky Linux

RHEL-compat

Stable, long-term RHEL alternative.

CentOS Stream

Rolling

RHEL development stream.

Fedora

Leading-edge

Latest packages and toolchains.

openSUSE

Coming soon

Not currently available. Planned to return in a future release.

Custom ISO

Upload

Boot from your own image via console.

Reinstall
Seconds from the panel or API
Console
VNC-style in-panel access
Root
Full root on every plan

Windows Server support is coming. We're starting with the Linux variants and will add Windows Server support at a later point. Linux-first.

Self-service & API

Deploy in minutes. Drive from your own tooling.

A web panel for humans, a documented REST API for everything else. Console access in-browser for emergencies. Use it in your Terraform, your CI or your own dashboard.

What you can drive from the panel API:

  • Rebuild
  • Reinstall
  • Reboot
  • Power on / off
  • Manage IPs
  • Console access
  • ISO upload
~/manage.sh

# Reboot a VPS via the panel API

$ curl -X POST https://api.buyvps.com/v1/instances

-H "Authorization: Bearer $TOKEN"

-d '{"region": "ams", "plan": "D-16", "image": "ubuntu-22"}'

{

"id": "vps-2k9x7p",

"status": "provisioning",

"region": "ams",

"plan": "D-16",

"ipv4": "203.0.113.42",

"ipv6": "2001:db8::42",

"ready_in": 38

}

Honest density

Overcommitment is the industry default. We refused.

For a long time now overcommit has been the industry standard to sell servers. Virtually all Virtual Private Servers are sold based off the sales chart for all of the vCPU and RAM that are available on servers, many of which are actually sold under capacity. Then when a customer buys more vCPU and RAM than a single physical node has available then only one customer at any time can be fully utilizing a customer's funds for a server. That customer will get low latency for packets and high performing applications until other customers start to take over the server. At that time the other customers will have to wait for their packets to arrive and their applications will perform poorly due to the increased competition for resources from the other customers on the server.

Typical VPS market

Oversubscribed
Node cap
CPU
150%
RAM
120%
  • Sold capacitySeller will be banking on their neighbors to remain idle, however the moment a customer ramps up their servers in serious heavy use the entire stack of oversubscription suffers.
  • Memory headroomNone by design
  • StrategyBanks on neighbors staying quiet

This hosting strategy works great for someone that wants to enable VPS hosting and need resources from several servers right away. However, as your neighboring VPS customers become active then your hosting will start to degrade significantly.

BuyVPS

Capped on purpose
Node cap
CPU
~80%
RAM
~68%
  • Sold capacityHard cap below physical
  • Memory headroomYes, unsold by design
  • StrategySized for peak, not average

Your share of a node's resources will always be there for you to use. That is your product. Your headroom is your product, not something that you can 'sell' and then run your applications on the remaining unsold 'hard capped' memory from the other resources on the server.

Node policy

Three node-policies. Three ways we promise never to overcommit.

As said above all 3 "node-policies" / "density" tiers are build with the same basis as for the memory (hard capped for the usable memory per node) but then differing designs for CPU sharing / vCPU "time-slicing" between the boxes as well as differing box placements in respect to each other on the respective tier.

Standard

Capped sharing

Shared vCPU on EPYC Milan Servers but designed with Low-Density (hard capped usable memory per node) to never be overcommitted so you have unsold headroom on your memory for your applications to grow.

vCPU shares per node1:2 to 1:4

Shared vCPU configuration. EPYC Milan nodes. Low-Density / hard capped memory for usable memory. Small amount of neighbors. Headroom unsold.

  • CPUShared vCPU, EPYC Milan
  • MemoryDDR4 ECC, capped per VPS
  • HeadroomYes, unsold headroom for your product
Best for: web apps, SaaS, dev/staging

Dedicated CPU

Reserved cores

These Reserved Cores are Pinned to your VPS and deliver the most predictable per-core latency under any load, the newest Genoa CPUs from EPYC paired with DDR5 ECC memory.

CoresPinned, 1:1

These Cores are reserved for you and as such will be pinned to your VPS in order to provide the most predictable per-core latency under load. Our newest CPUs, EPYC Genoa.

  • CPUReserved core(s), latest Genoa EPYC processor
  • MemoryDDR5 ECC, capped per node
  • HeadroomBy design, all reserved cores are never shared with neighbors
Best for: databases, APIs, CPU-bound apps

High Memory

Memory-first

Large memory pool for in memory workloads. The CPU on these High Memory Node are the same Milan CPUs as the Standard Nodes are running on. The configuration is however Memory-cache sized for the number of tenants to run on these High Memory Nodes. Large memory headroom was what these were designed for.

RAM : vCPU8 : 1

In-Memory workloads are built to run within a large pool of memory and are sized within a node for in-Memory headroom within the memory cache given the tenant count for that node.

  • CPUShared vCPU, EPYC Milan
  • MemoryDDR4 ECC protected
  • HeadroomMemory cache sized for tenant count
Best for: databases, caching, analytics
What you sign up for

Commitments we don't walk back.

No teaser pricing. No setup fee. No retention scripts.

Hardware

NVMe RAID10 and ECC memory on every node

Mirrored and striped storage with error-correcting memory.

Price

No setup fee

You pay the plan price, with nothing added at the start.

Term

Cancel any time

All from the panel. No retention script required.

Support

Engineers reply

Direct from the engineers (and not through a chatbot), no deflection to a tier-1 support person.

30-day performance guarantee. Not satisfied? Full refund, no admin fee. How it works

FAQ

Common questions

Do you use KVM virtualization?
Yes. Every VPS runs on KVM with hardware-assisted virtualization and its own kernel, so you get full root access. CPU, memory and disk are allocated and isolated per instance at the hypervisor level, with no container-based sharing like OpenVZ.
What does low-density node policy mean?
We cap usable memory per node instead of filling each server to its theoretical maximum. Fewer instances per node means less contention for CPU and I/O, and more consistent performance under load.
What hardware do you use?
AMD EPYC single-socket servers. Milan on Standard and High Memory, Genoa on Dedicated CPU. DDR4 ECC on Standard and High Memory, DDR5 ECC on Dedicated CPU. NVMe RAID10 storage. Power and network paths are redundant to each rack.
How fast is the storage?
All plans use NVMe storage. We measure 4K random I/O, the access pattern real workloads generate, rather than only sequential throughput.
Do you publish benchmarks?
Yes. CPU, disk I/O and network results are published per location and per plan. Full test configuration and methodology are published on buyvps.com/vps-benchmarks/.
Where can I deploy?
Amsterdam (Netherlands) and New York (United States).
Can I upgrade without migration?
Yes. You can add RAM and CPU in place, keeping your data and IP address. Most upgrades need a single short reboot to take effect.
How does pricing work?
You pay the price of the plan you pick, every billing cycle. You can upgrade or downgrade. Any one-time fees are listed before checkout.
Keep exploring
See VPS plans and pricingFind the right fit for your workload.Reproducible benchmarksThe numbers behind the hardware.Where we hostAmsterdam and New York data centers.Supported Linux distributionsPlus custom ISO upload.