The table shows only the cheapest matching plan per provider. With no limit, the winner is often burstable/Lightsail without a Mark score (–). Raise the slider → AWS EC2 / GCP AMD / Azure D&GPU / Oracle with PassMark enter the comparison.
Single is comparable to the Single column on dedicated servers. CPU Mark (scaled ⚠︎) is not directly comparable to a full-chip dedicated server.
| Provider | Plan | CPU | vCPU | RAM | Disk | GPU | Traffic | Egress €/GB | Actual €/mo | CPU Mark | Single | €/1000 pts | Datacenter | Trustpilot | Notes |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Loading… | |||||||||||||||
How are VPS performance measurements made?
CPU is measured with sysbench using one thread and all visible threads. Disk is measured with fio direct I/O: sequential 1 MiB read/write plus random 4 KiB read/write, including p99 latency. Each displayed result is a median and includes the observed minimum–maximum range and measurement count. Compare only results made with this same method. A shared VPS can vary by host load, so one run is not treated as a stable rating. Internet speed is not shown until a controlled iperf3 endpoint is available; an arbitrary speed-test server would mix provider performance with route and remote-server load.
How does search work?
We fetch each provider’s full public VPS catalog automatically once a day, but the table shows only the cheapest matching plan per provider (e.g. AWS Lightsail and EC2 compete on the same row). Prices are normalized to euros excl. VAT (consumer price with incl. VAT selected); USD-billed prices use the ECB daily rate and are unaffected by the VAT toggle. Plans are not identical — alongside price, check disk type and traffic columns (∞ = unlimited, – = not stated; missing data fails filters) and commitment terms in Notes. Source is linked from the provider name. UpCloud and Contabo prices are checked manually because their sites block automated fetching. CPU Mark (AWS EC2, GCP AMD, Azure GPU host, Oracle E4, UpCloud) is PassMark’s multi-thread score as a scaled estimate (chip score × vCPU / threads) — marked ⚠︎. UpCloud Premium = EPYC 9575F; Starter ≈ EPYC 7542 (fleet sometimes 7543). Single is PassMark single-thread — same scale as dedicated servers, so per-thread quality is comparable. Burstable and unknown chips = –. Use Min CPU Mark / Min Single sliders to compare performance, not just the cheapest burstable. €/1000 pts = monthly price / (CPU Mark / 1000); lower is better value.
Actual cost, not list price
List price is not what you pay. The usage profile controls (outbound egress, number of public IPv4 addresses) compute actual €/mo and sort by cheapest actual cost — not headline price. Public IPv4 is added for providers that charge separately (e.g. Hetzner and Scaleway bill even the first address), and egress is calculated only on traffic above the plan’s included allowance — 0 for unmetered or unlimited providers. IPv4 and egress figures come from IPv4 and egress comparisons; if extra cost is unverified, it is marked ? and excluded from the sum — actual cost may then be higher than shown. We never invent numbers.
Hyperscaler hidden costs this comparison does not include
AWS, Azure and Google Cloud bills often carry line items no instance comparison will show. The usual surprises: NAT Gateway (AWS ~$0.045/h + $0.045/GB processing — its mere existence costs ~€35/mo, and private-subnet machines need it for outbound access), CloudWatch logs ($0.50/GB ingestion — a chatty app produces tens of GB a month), cross-AZ traffic ($0.01/GB each direction — hits exactly the HA setups where replication crosses zones), S3 API request fees and Glacier retrievals, RDS backups beyond the free quota, and Business Support (~10% of the whole bill). European flat-price providers (Hetzner, UpCloud) mostly have none of these — the flat monthly price includes basic networking.
Commitment discounts work in the other direction: hyperscalers sell below list price if you commit — AWS Savings Plans / Reserved Instances (1–3 yr, typically 30–60% off), Azure Reservations and Google Cloud Committed Use Discounts. OVH's from-prices already require a 12–24 month commitment (noted in the table). Prices in this comparison are on-demand list prices with no commitment — that keeps rows comparable, but remember a committed hyperscaler price can be a third cheaper, while flat-price providers' list price already is the final price.