Unmetered says the bytes are not counted; it does not say the port is yours at the busy hour. The five contract clauses to check, how to size a tier from your peak, what each tier costs per delivered terabyte, and the 13 cities where the server can live at the same price.
An unmetered dedicated server is a whole physical machine whose monthly price covers a network port rather than a quantity of traffic. You pay for the size of the pipe; what flows through it does not appear on the invoice. That one sentence hides most of the difference between a good offer and a bad one, because "unmetered" says nothing about whether the pipe is actually yours at the busy hour. This guide covers what to check in the contract, how to choose a bandwidth tier, what each tier really costs per terabyte, and where you can put the machine.
Hosting pages use these words loosely, so pin them down before comparing prices:
The trap is "unmetered" without "guaranteed". A 1 Gbps port on a switch whose uplink is shared twenty ways is unmetered, and it may deliver 150 Mbps at 21:00. What unmetered bandwidth actually means covers how that oversubscription works and how to spot it.
The plan page sells the port. The terms decide whether you keep it. Ask for these in writing:
Here is how our own dedicated plans answer them. The listed rate is guaranteed in each direction, and using all of it around the clock carries no cap and no overage charge. Short bursts above the guaranteed rate are allowed; if a server runs above it repeatedly, our terms say we will contact you about moving up a tier — refusing can end in suspension, but past overuse is never billed. And the port can be raised to a higher tier on the same machine rather than by migrating to a new one.
Start from the traffic, not the hardware. One gigabit per second held flat out for 30 days moves:
1 Gbps = 125 MB/s × 86,400 s × 30 days ≈ 324 TB a month
No real service runs flat out, so divide by your utilisation. A download or video origin averaging 40% of its port uses about 130 TB a month per gigabit; a busy evening-peak service might average 25%. Work out your peak rate, size the port for that, and the monthly total takes care of itself. Choosing an uplink speed has worked examples for streaming, file hosting and VPN pools.
Rough starting points:
Unmetered pricing only makes sense against what the same traffic would cost metered. Here is the entry price of each ZetServers tier, with the cost of a delivered terabyte at full use and at a realistic 40%:
| Tier | From | TB/month at 100% | €/TB at 100% | €/TB at 40% |
|---|---|---|---|---|
| 1 Gbps | €209 | 324 | €0.65 | €1.61 |
| 2 Gbps | €259 | 648 | €0.40 | €1.00 |
| 5 Gbps | €519 | 1,620 | €0.32 | €0.80 |
| 10 Gbps | €659 | 3,240 | €0.20 | €0.51 |
| 20 Gbps | €1,159 | 6,480 | €0.18 | €0.45 |
| 50 Gbps | €2,799 | 16,200 | €0.17 | €0.43 |
| 100 Gbps | €4,499 | 32,400 | €0.14 | €0.35 |
Two things stand out. The price per terabyte falls steeply up to 10 Gbps and then flattens, so the jump from 1 to 10 Gbps is the one that changes the economics. And utilisation matters as much as the tier: a 10 Gbps port you fill to 40% costs less per terabyte than a 1 Gbps port you fill completely. The cost per terabyte of bandwidth shows how to run the same sum on any quote.
The "from" prices use the entry configuration of each tier — a dual Xeon E5-2630 v4 up to 20 Gbps, a dual Xeon Gold 6148 from 25 to 100 Gbps — and the hardware rises from there, to dual EPYC at the top. Bandwidth and hardware are chosen separately, so a modest CPU on a large port is a valid order if the workload is delivery.
Our unmetered dedicated servers are available in 13 cities:
| Region | Cities |
|---|---|
| Central Europe | Frankfurt, Amsterdam, Vienna, Warsaw |
| Western Europe | Paris, Dublin |
| Southern Europe | Madrid, Lisbon, Milan, Marseille |
| South-east Europe | Bucharest, Sofia |
| North America | Ashburn |
The location does not change the price. In our order system the city is a free option on every dedicated plan, so a 10 Gbps server costs the same in Lisbon as in Frankfurt. What differs between cities is stock — a particular configuration may be ready in one city and built to order in another — and who the city is close to.
That second point is where the choice should be made. Each city sits nearest a different set of users and networks:
The next six articles in this series take one city each and work through who it serves, what the latency floor looks like from there, and which workloads suit it.
It is not always the cheaper option. If your server moves 3 TB a month — a typical business website, an internal tool, a small database — a metered plan with a 20 TB allowance is cheaper and you will never touch the limit. Unmetered pays off when traffic is heavy, steady or unpredictable: delivery, streaming, VPN, mirrors, backup targets, anything where a bill that scales with success is a problem.
The traffic volume is unlimited; the rate is not. You can transfer as much as the port carries, which at 1 Gbps is about 324 TB in a 30-day month. Beyond that the limit is physics, not a bill.
Unmetered means traffic is not counted. Guaranteed means the port rate is reserved for you rather than shared. You want both: an unmetered port on an oversold uplink can still be slow at the busiest hour.
Yes. Bandwidth is chosen separately from the hardware, so moving up a tier is a change to the port, not a migration. The very top tiers need a matching network card, so mention 100 Gbps plans when you order. Plan for the peak you expect within a year and grow from there.
No. The city is a free option on our dedicated plans, so the same configuration costs the same in all 13 locations. Availability differs: some builds are in stock in one city and assembled to order in another.
Most in-stock configurations are active within an hour of confirmed payment, and an operating system install takes 15 to 20 minutes after that. Rarer builds can take two to four weeks; ask on live chat before ordering.
Generating a token means reading the model's active weights once, so a CPU server's speed ceiling is its memory bandwidth divided by the model's size. The arithmetic, the catch in prompt processing, why two sockets are not twice as fast, and the workloads a big-memory EPYC box genuinely suits.
Most database performance problems are a working set that no longer fits in memory, or a commit that waits on a slow flush. How to measure both, how much RAM and what kind of disk to buy, where the CPU matters, and how to lay out replicas and backups.
One rented machine, a free hypervisor, and as many virtual machines as the hardware will hold. How to budget RAM, lay out ZFS, get networking right on a data-centre switch port — the part that differs from a homelab — and keep the backups off the box.