Skip to content
Getting Digital

Dedicated Server

Also: dedicated hosting, bare metal server

A dedicated server is an entire physical computer rented by a single customer, with no hypervisor taking a share and no other tenants on the hardware, chosen when consistent performance, specific components or provable isolation matter more than the ability to resize in seconds.

Our take. Nearly everybody convinced they need one actually needs a bigger VPS and a cache that works. The genuine cases are narrow and identifiable: a database that must not share a disk with strangers, software licensed per physical processor core, or an auditor who wants single tenancy on paper. Buy it for one of those reasons, not because owning a whole computer sounds more professional.

One tenant, one machine

What arrives when you rent bare metal is a particular computer in a particular rack: a named processor, a fixed quantity of error-correcting memory, disks you selected, a network port, and out-of-band management through IPMI or a vendor console so you can reinstall the operating system or watch it start up without anybody walking over to it. No hypervisor takes its cut, and no neighbour's overnight backup job turns into your slow query. That pitch still holds for a narrow set of workloads. Databases care about consistent disk latency far more than about good average latency. Commercial software licensed per physical core can cost more on a virtualised host than the hardware underneath it does. Regulated environments sometimes need single tenancy written down rather than argued about. And the arithmetic favours a whole machine for anything genuinely busy all day, because you pay for the full box either way and a box running near capacity is cheap per unit of work done. What you surrender is everything the cloud normalised. Capacity appears when the provider racks it, not while you wait on the phone. Growing means ordering a second machine and building the coordination between the pair yourself. And hardware breaks: when a disk dies you suddenly care a great deal whether you configured redundancy, whether a spare is already in the chassis, and how fast somebody on site will swap it, because a virtual machine can be rebuilt elsewhere in minutes while a physical one has to be repaired where it stands. Providers have blurred the boundary with bare-metal products that provision real machines through an API and bill by the hour, but the underlying trade has not moved an inch.

Ask these before you order

Both answers live in the provider's own documentation and neither is on the specification sheet. First, what happens when a disk fails: who notices it, who replaces it, inside what window, and whether the replacement is billed. Second, how you reach the machine when it will not boot, which is what remote console access is for and why a plan lacking one is really a plan to open a ticket and wait. Single tenancy removes the noisy neighbour, but it also removes the platform that used to quietly migrate your workload off dying hardware while you slept. On your own iron that work belongs to you, and it is made of disk redundancy, restored-and-verified backups, and monitoring loud enough to wake somebody.

In practice

Hetzner's server auction shows what bare metal really is better than any brochure: a live inventory of machines already racked in their facilities, each listed with its specific processor, memory, disk set and location, available immediately at a price that falls the longer the unit sits unsold. Nothing is abstracted away. You are not picking a size from a menu of labels, you are picking a computer, and when its disks fill up there is no slider to drag. Compare that with choosing an instance type on a cloud platform, where the label describes a performance class and the silicon behind it may differ between one reboot and the next. The auction also makes the economics legible: hardware a generation or two old, with plenty of cores and memory, is priced as stock to be cleared, so plenty of it ends up quietly running build agents and self-hosted services.

Often confused with

VPS (Virtual Private Server)
Machines like this one are carved into portions and sold by the portion; that product is a virtual private server. Dedicated simply means no portion of yours belongs to anyone else.
Cloud Hosting
Cloud capacity is billed for the minutes it exists and can be replaced in software; bare metal is a fixed monthly line on the invoice and, when something breaks, a physical repair.

Key takeaways

  • →Physical exclusivity buys the most consistent performance available and the strongest isolation, at the highest fixed price of any hosting model.
  • →It suits steady, heavy or compliance-bound workloads precisely because it gives up per-hour billing and instant resizing.
  • →Component failure becomes a scheduling problem rather than an abstraction: redundancy, tested restores and remote console access are not optional extras here.

Related concepts

More courses from these shelves

A rotating selection from the course directory, drawn from the subcategories where this concept is taught rather than picked for it. Details, price and the provider link are on the course page.

Introduction to Cloud Data Analytics with Google BigQuery

This course is designed for the students who are at their initial stage or at the beginner level in learning the data w…

Udemy

Begin with Microservices

In this course, you will learn how to build REST APIs or Microservices using the latest version of Spring Boot), and Mo…

Udemy

AWS EKS (Elastic Kubernetes Service)

The Course is divided into 3 parts -Part 1 - Kubernetes and its objects (Deep Dive). In this part I will talk about Kub…

Udemy

Increasing BDD Code Efficiency

Increasing BDD (Behavior Driven Development) Code Efficiency is all about minimizing your maintenance effort while maxi…

Udemy

Azure Kubernetes Service AGIC Ingress: 30 Real-World Demos

Course Overview Welcome to this Amazing course on Azure Kubernetes Service AGIC Ingress: 30 Real-World Demos. Below is…

Udemy

Monitoring and Maintaining Agent Performance

Are you building, deploying, or managing AI agents and want to ensure they operate at peak performance? Monitoring and…

Udemy

FAQ

Who genuinely still needs one?
Anyone whose demand is constant enough that paying for a whole computer beats renting elastic capacity, plus anyone whose constraint is not performance at all. High-traffic shops, latency-sensitive databases, game servers, software priced per physical core and organisations whose rules forbid shared hardware are the recurring cases.
Is it faster than a virtual server?
At equivalent stated specifications, usually yes and reliably more consistent, because no hypervisor sits in the path and nobody else is competing for the same disks and network card. Whether that gap justifies the price difference is a calculation per workload, and for a site that spends its day waiting on an uncached database it often does not.
Managed or unmanaged?
Unmanaged means the provider guarantees the hardware and the network connection and nothing above it, so operating system updates, security hardening, backups and diagnosis are yours. Managed variants add an operations team and a considerable fee. Choose by asking who will be awake and competent when the machine stops answering, because on bare metal that question has a physical answer.

Sources

The primary text this definition rests on. Read it before you trust ours.

Last reviewed 14 September 2026 · Getting Digital