About the project

VPS Observatory is a research project about the real performance of virtual servers.

We take VPS instances from providers, run the exact same set of tests on each of them and publish everything we measured: CPU, memory, disk, network, system facts and the measurement history. No marketing retold.

Choosing a VPS today works strangely. A provider writes “2 vCPU, 4 GB, NVMe, up to 1 Gbit/s” — and that describes the invoice, not what you will get. A plan line says nothing about the CPU generation, how many neighbours share your disk, or what “up to 1 Gbit/s” means under load.

There is nowhere to compare: reviews retell the plans, testimonials are emotional and often commissioned, and rankings are usually for sale. Checking yourself means renting a dozen servers and spending a week.

We spend that week for you and show not an opinion, but numbers together with the method behind them.

An agent connects to the server over SSH, installs standard open-source tools (sysbench, fio, openssl, curl) and runs an identical measurement programme. Key tests are repeated, the median is used, and the spread across passes is published separately — it says as much about stability as raw speed does.

The metrics are then normalized and folded into 0–100 scores using an open, versioned formula.

The full methodology: tools, weights, formula and its limits →

Tests cost money: servers must be rented, and traffic and time paid for. The project earns from advertising and affiliate programmes, and we say so up front rather than hiding it in the footer.

• Ad blocks are labelled and visually separated from rankings.
• A click through to a provider’s site may use an affiliate link — in which case we earn a commission. Your price does not change.
• Neither advertising nor affiliate payouts change measurements, scores or positions. The formula is open — you can verify this rather than take it on faith.

Advertising decides whose block you see at the top of a page. It does not decide what numbers a test reports.

The project is run by an engineer who has spent years working with hosting infrastructure: servers, virtualization, networks, bandwidth shaping, and debugging outages at three in the morning.

Hence the approach. Someone who has operated nodes themselves knows that “up to 1 Gbit/s” and a steady gigabit are different things, that random IOPS matter more than pretty sequential megabytes, and that spread across passes tells you about your neighbours on the node. So this project measures what actually hurts in production, not what looks good in an ad.

The measurement base keeps growing: the more tests across locations, the more reliable the comparison and the more accurate the normalization. Planned: more providers and countries, regular re-measurements of the same plans (to reveal degradation over time), and a more detailed network picture across routes.

If you are a provider and want to be measured — write to us. There is one condition: the result gets published as is, whatever it turns out to be.