Benchmarked faster than DigitalOcean, Vultr & OVH. Your price never rises at renewal. See the numbers Benchmarked #1. No renewal hikes. See the numbers
Skip to content
Sign up
NVMe VPS

NVMe VPS, measured not marketed.

All plans run on NVMe RAID10 storage, individual fio benchmark results for each region. No SATA SSDs, no SAN, no "up to" numbers. 122,959 read IOPS fio median for D-4 (smallest Dedicated plan for D-4).

Compare every plan View raw benchmarks
Why NVMe

The interface matters more than the medium.

Why NVMe. The interface matters more than the medium. NVMe is the modern interface between an SSD and a CPU. SATA was designed for spinning disks (hard drives), NVMe was designed for flash storage (SSDs). So the key performance metrics are random small I/Os (IOPS and write latency, both very important for database workloads as well as for concurrent web workloads.

01

NVMe vs SATA SSD

SATA SSDs are limited by the older interface, whereas NVMe SSDs are able to handle significantly higher IOPS due to the nature of the interface. SATA SSDs are generally limited to around 100k IOPS (theoretical), whereas NVMe SSDs are able to reach and exceed 500k IOPS.

02

RAID10 underneath

The Storage within our nodes are comprised of NVMe Storage Drives in a RAID10 configuration to ensure the best possible performance, and also to provide redundancy for individual Storage Drives. Meaning that a single Storage Drive can fail, but all storage on that individual drive will be shifted to other Storage Drives without any issues.

03

Local, not SAN

Our servers use local storage. The storage is on the local server and there is no SAN in between. So, this setup has lower latency, less failure points and more predictable performance.

04

Measured per plan

fio benchmark results per plan and region including raw logs for verification.

Hard numbers

NVMe RAID10 in fio.

We provide hard numbers for our plans. These are the median values of 5 fio runs for a D-4 plan (2 vCPU Dedicated, 4 GB RAM, NVMe RAID10) in Amsterdam.

122,959
4K random read IOPS. fio is run with 32 queue depth and direct I/O to mimic a database load.
87,730
4K random write IOPS. Sustained write performance under concurrent load.
693.84 MB/s
fio mixed 4K throughput. 50/50 read/write. The realistic shape of database I/O.
+37%
Above DigitalOcean baseline. We have run the exact same test on their own hardware, using the same configuration (i.e. the same plan size, same fio options, etc.) as the test we ran on DigitalOcean to benchmark them.
Full fio methodology, configuration files and raw logs
Where IOPS matter

Workloads that benefit from NVMe RAID10.

OLTP databases

Small random read & writes, many of OLTP databases are based on PostgreSQL, MySQL, MariaDB.

Document stores

MongoDB and similar database engines. Many concurrent random I/Os across many collections of a database. IOPS budget is what usually limits the database's throughput.

Search indexes

The rebuild of search indexes as well as the concurrent I/O for queries of search indexes (Elasticsearch / OpenSearch / Meilisearch) also benefit from low write latency.

High-concurrency web apps

App servers with many parallel requests doing small file or cache reads. Tail latency drops when the storage gets out of the way.

Container hosts

Docker and k3s hosts. Many small read operations of layered images of your applications. Many small write operations of your containers' log files. The I/O load of image builds are processed extremely fast with NVMe storage.

CI runners

git clone, npm install, pip install, cargo build. Faster random I/O directly cuts CI minutes.

Pricing

NVMe RAID10 on every plan.
No tier upgrade required.

NVMe RAID10 is not a feature that can be unlocked at the top tier, it is the architecture of the storage on every plan from S-4 to H-192. Therefore, the only thing that needs to be chosen is the plan that best fits other requirements (CPU, RAM, etc.).

Compare every plan
Standard$18.40/moShared 1:4 vCPU, S-4 to S-64
Dedicated CPU$29.60/moPinned 1:1 cores, D-4 to D-64
High Memory$199.20/mo1:8 RAM ratio, H-64 to H-192

NVMe RAID10 from the smallest plan up.
Same storage, every plan.

Choose a plan and a region. The storage of every plan is identical for sizes S-4 through H-192.

Deploy a VPS Full benchmarks
FAQ

Common questions

Is NVMe RAID10 available on every plan?
All plans from S-4 through H-192 run on NVMe RAID10 storage. NVMe is not a premium add-on feature; it is the standard storage architecture on every plan.
How is NVMe different from SATA SSD?
SATA storage was originally designed with spinning disk storage in mind and has a ceiling of around 600 MB/s with very limited IOPS (input/output operations per second) per disk. NVMe storage on the other hand was specifically designed for flash storage (SSD) and can achieve 500k+ IOPS per individual drive. For random small-block I/O workloads (for example databases and search indexes), NVMe-based storage will typically offer a huge increase in performance over equivalent SATA-based storage.
What does RAID10 protect against?
With RAID10, even the failure of a single drive is not a problem for your storage, since data from each RAID10 member drive is striped and mirrored across more than one physical drive. You get the read performance of striping plus the redundancy of mirroring.
How are the IOPS benchmarks measured?
For our IOPS benchmarks we used fio on each server local storage, at a queue depth of 32, using direct I/O (no file system) and 4K block size, with a fully random read and write pattern. For each test, we ran 5 trials and report the median value. You can find the full test configuration and raw test logs for each test on the benchmarks page.
What workloads see the biggest gain from NVMe RAID10?
OLTP databases (PostgreSQL, MySQL), document stores (MongoDB), search indexes (Elasticsearch), high-concurrency web apps, container hosts (Docker, k3s) and CI runners.
Is the storage local or network-attached?
All of our storage is local to the physical server (no SAN, no network-attached storage). It is lower latency, fewer failure points and more predictable performance.
What happens to my data if a drive fails?
The storage on our servers for your VPS is set up as RAID10, meaning that if a drive were to fail for any reason the other drives in the mirror would continue to run your VPS without issue. We would then swap the failed drive during scheduled maintenance and the RAID would rebuild your data from the good drive in the background.
Can I add more NVMe storage to my plan?
Extra NVMe storage is offered as an add-on in 100 GB, 250 GB or 500 GB increments. The cost for extra NVMe storage for Standard and High Memory plans is set on a single price track. For Dedicated CPU plans, extra storage is offered on a separate price track.
How does the IOPS scale with plan size?
Each VPS has an IOPS ceiling determined by the VPS plan and the number of vCPUs the VPS has. A VPS in a lower-end plan will have a lower IOPS ceiling than a VPS in a higher-end plan. Concrete numbers per plan are published on the benchmarks page at buyvps.com/vps-benchmarks/.
What does no overselling mean for storage?
The storage which is reserved for a plan by the hypervisor on the underlying NVMe RAID10 is the storage that is shown for a plan. There is no overselling of storage. We do not count on our customers not fully using the allocated storage.
Ready to deploy

Deploy a measured VPS

From $18.40/mo on a 2-year term. KVM on AMD EPYC, NVMe RAID10, ECC memory. Your price is fixed for the lifetime of the service.

See all plansDeploy now
Keep exploring
Compare all VPS plansNVMe RAID10 on every plan.Storage IOPS benchmarks4K random I/O, measured.The storage stackNVMe RAID10 with PCIe direct.