How shared 1 Gbps ports work in VPS hosting

Almost every VPS on the market — ours, Hetzner's, OVH's, DigitalOcean's, Vultr's — is advertised with a 1 Gbps network port. Almost nobody clarifies what "1 Gbps" really refers to. This article does.

The short version

1 Gbps is the speed your VPS can reach in short bursts when the physical uplink is otherwise idle. It is not a private, dedicated lane that belongs only to you.

Each physical server in our datacenter has one (or more) 1 Gbps uplink. That uplink is shared between every VM running on the box. Hundreds of customers, one pipe. The same arithmetic applies at every other major hosting provider.

Why the math works

It works because customer traffic is incredibly bursty. A typical workload — a website, a small game backend, a Telegram bot, a personal Mastodon instance — sits at near-zero traffic 99% of the time and spikes briefly when something happens. If we measured every VPS on a node for a week, the average load would rarely exceed a few Mbps per VM.

Selling "100 Mbps dedicated" instead of "1 Gbps shared" would mean throwing away 90% of the available capacity to satisfy customers who would never use it anyway. Nobody in the industry does that, because the numbers don't add up — a true dedicated 1 Gbps port from a tier-1 carrier costs €500–1500/month. Selling that for €5 would be a one-way ticket to bankruptcy.

What this means for your VPS

• You can hit close to 1 Gbps in a short burst (say, downloading a Linux ISO) — when nobody else on the node happens to be transferring at the same moment.
• Sustained throughput depends on the hour of day, your neighbours on the node, the route to your destination, and your destination's own capacity.
• Single-flow speeds (one TCP connection from one source to one destination) are usually limited by the route, not by us. Two parallel transfers will often add up to more than one.
• Speed-test tools (speedtest.net, iperf3 to a random host on the internet) are not a measurement of "your VPS" — they measure the entire path, every hop, every overloaded transit link in between. Deeper version — VPN overhead, distance/RTT effects, the real cost of dedicated bandwidth, and how to test properly — is in our explainer on VPS internet speed.

Fair use, in plain English

We don't enforce hard rate limits on entry tariffs. We do reserve the right to throttle a customer who runs sustained traffic that meaningfully degrades service for everyone else on the node — for example, a public file mirror pushing 400 Mbps for 12 hours straight.

If you actually need that kind of sustained throughput — a CDN edge, a public mirror, a streaming origin, a high-traffic proxy — that's a legitimate workload, but it doesn't fit a €2 entry tariff. Open a ticket and we'll quote you a plan with dedicated bandwidth.

How to get a real, predictable picture

1. Test from inside the VM with iperf3 against a public iperf endpoint located in the same datacenter region — that isolates your VM's network from everything that happens in transit.
2. Test at different times of day. Bandwidth on a shared resource is statistical, not constant.
3. If you need a guaranteed number — write to support. We'll either confirm whether your plan is the right fit or move you to one that is.

If you want the full picture of why an end-to-end speedtest from your laptop in another country shows what it shows — public internet vs our fiber, VPN-tunnel overhead, the Mathis throughput limit, real prices of dedicated point-to-point bandwidth — read the companion article Internet speed on a VPS, honestly explained.