Benchmark Methodology
What each column on our benchmarks table actually measures — and, just as importantly, what it does not.
Scope and limits
Read this part first. It is the honest shape of the data.
- • Our benchmark coverage is partial. We hold results for a small number of providers and plans, not for the full catalogue. A provider with no row on the table has not been benchmarked by us.
- • Every result is a snapshot from the date shown on its row. We do not currently re-run benchmarks on a fixed schedule, so check the date before relying on a figure.
- • Results are tied to a specific plan and region. A score for one plan does not transfer to another plan from the same provider, and shared-CPU plans vary between neighbours on the same host.
- • Not every figure is a measurement we made. The “measured by” column on each row says who produced it, and the sections below say which columns are measurements and which are provider-stated capacities.
CPU — sysbench
CPU scores come from the sysbench cpu benchmark, run on an instance provisioned for the test. It measures integer computation throughput, and higher is better. It is a synthetic CPU test: it tells you how fast the vCPUs crunch arithmetic, not how your application will behave.
Disk I/O — fio
Disk figures come from fio: sequential read and write throughput in MB/s, and random 4K IOPS. For database and other metadata-heavy workloads, the IOPS figure is usually the one that matters; for bulk file movement, the sequential figures are. Measurements are taken on the default storage tier the provider ships with that plan, so an NVMe plan and a SATA SSD plan are not like-for-like.
Network capacity
The inbound and outbound Gbps figures we hold are the network capacity the provider states for that plan — the port speed or egress allowance from their documentation. They are not a throughput test we ran, and real sustained throughput is usually lower than the stated ceiling. We do not currently publish measured network throughput.
Latency
Where a latency figure is present it is an HTTP response-time sample taken from the instance against a static file in the same region, recorded in milliseconds. It is a single-region sample, not a global latency matrix, and it will be dominated by where the datacentre is relative to the thing you are talking to.
Who measured it
- • ServerSpotter — benchmarks we ran ourselves on an instance we provisioned.
- • cloudperf.run — published benchmark data from cloudperf.run.
- • vpsbenchmarks.com — community-submitted results from vpsbenchmarks.com.
Rows we did not measure are attributed on the table and linked back to their source so you can check the figure where it was published. Third-party results were not produced under our conditions, so treat cross-source comparisons with more caution than comparisons between two of our own runs.
How this relates to our ratings
Benchmark results do not feed a scoring formula. Our provider ratings are editorial — set and reviewed by the ServerSpotter team — and benchmark figures are one input the team weighs alongside specs, pricing and a provider's public track record. See how we test for the full picture, including how we handle paid placement.
Corrections
If you are a provider and you believe a result here misrepresents your platform, or you spot something that looks wrong, tell us and we will re-check it.