VPS vs dedicated server: which to rent
What really separates a shared hypervisor from a whole machine: noisy neighbours, steal time, price steps, and when each one is the cheaper buy.
The short answer
A VPS and a dedicated server differ in one thing: how much of a physical machine you get. A VPS is a virtual machine on a host that runs other virtual machines beside yours, so the CPU, the disk and the network card are shared under a hypervisor. A dedicated server is the whole physical box, with nothing else on it. Everything else people argue about, performance, price, isolation, follows from that single fact. Most readers should rent a VPS, because most workloads are bursty and a VPS can be resized in minutes. Rent dedicated hardware when your load is steady and saturated, when a rule says the hardware must be yours alone, or when you need more memory than any virtual plan offers.
What actually differs: a shared hypervisor versus a whole machine
On a VPS, a hypervisor (the software that runs virtual machines on one physical host) hands your guest a slice of the real hardware. You get virtual CPU cores mapped onto physical cores, a share of the host's memory, and storage carved out of the host's drives. The hypervisor enforces the boundary, so another customer on the same host cannot read your memory or your disk. What it does not fully enforce is speed. Physical cores, disk queues and the network uplink are finite, and they are shared.
On a dedicated server there is no hypervisor between you and the hardware, unless you install one yourself. Every core, every memory channel, every drive and the full network port belong to one tenant. Nobody else competes for them, because nobody else is there.
You can see which side you are on in one command:
systemd-detect-virtIt prints kvm on a typical virtualised plan, lxc on container virtualisation, and none on bare metal. lscpu | grep -i hypervisor tells the same story from the CPU flags. That check matters for more than curiosity: some software needs hardware virtualisation extensions, which is why running virtual machines inside a VPS depends on the plan type you bought.
Noisy neighbours, and how to measure one
A noisy neighbour is another guest on your host using so much of a shared resource that your guest waits. The effect is real, and it is measurable rather than something you have to guess at.
The number to watch is steal time: the percentage of time your virtual CPU was ready to run but the hypervisor gave the physical core to somebody else. It appears as the st column in vmstat and the %st field in top.
vmstat 1 5Steal at or near 0 means you are getting the CPU you paid for. A few percent under load is normal on a shared plan. Sustained double-digit steal means the host is oversubscribed and your application is waiting on hardware it cannot see. That one measurement turns a vague complaint into a support ticket worth filing, or into a reason to move.
Disk is the other shared resource, and for databases and busy web applications it usually matters more than CPU. A host that puts every guest on the same drives shares those input and output operations too, so the drive technology behind the plan changes what you feel: how NVMe compares with SATA SSD on a VPS shows what that difference looks like in practice.
A dedicated server removes noisy neighbours completely. It does not remove contention. Your own services still compete with each other, and a runaway backup job will slow your own database just as effectively as any stranger.
The price steps
Virtual plans move in small steps. You add memory, cores or disk in increments, and the bill moves by a similar increment. Dedicated hardware moves in one large step, because you are renting a whole machine with whatever it happens to contain, plus the rack space, power and hands that keep it alive.
As of July 2026 that step is still close to an order of magnitude: a generous VPS rents for a few dollars a month, while entry level dedicated hardware usually starts in the high tens of dollars a month and climbs from there. Check current pricing before you plan around any figure, because both sides move.
The step has a second cost people forget: time. A VPS resize is a control panel action and a reboot. Moving to a bigger dedicated machine is a migration, because the disks are physically in the old box. Rent hardware you will grow into, or accept a migration later.
When a dedicated server genuinely wins
Steady, saturated CPU. If your load average sits near your core count all day, every day, you are paying for capacity you actually use, and shared plans are priced on the assumption that most guests are idle most of the time. Video encoding, continuous builds, simulation and large batch jobs live here.
Strict compliance or contractual isolation. Some audits, some customers and some regulations require that no other tenant occupies the hardware. That is a requirement about the machine, not about your configuration, so no amount of hardening on a shared host satisfies it.
Very large memory. Virtual plans stop at whatever the provider offers. When you need hundreds of gigabytes of RAM for an in memory database or a large model, dedicated hardware is often the only way to buy it.
Predictable disk behaviour at high volume. A busy database that must not see a latency spike benefits from drives nobody else is queueing against.
Hardware you must choose yourself. A specific CPU generation, a GPU, or a particular disk layout is a bare metal conversation.
When a big VPS is the smarter buy
Bursty load, which is most load. A web application that is busy for three hours and idle for twenty one is exactly what shared infrastructure is priced for. You buy headroom for the peak without paying for a machine that idles through the night.
You are still sizing the workload. Guessing wrong on a VPS costs a resize. Guessing wrong on dedicated hardware costs a migration, or a year of paying for the wrong box.
You want snapshots and fast rebuilds. Virtualisation gives you image based snapshots, cloning and rebuilds in minutes, because your disk is a file on the host. On bare metal, recovery means reinstalling or restoring from backup.
You want more than one machine. Two or three small servers usually beat one large one for resilience, and splitting roles across boxes is easier to do at virtual prices. Whichever you rent, the first hour of work is the same: the first ten minutes on a new server covers the users, keys and firewall that should exist before anything is published.
How to decide without guessing
Measure the machine you already have. Run vmstat 1 60 through your busiest hour and read the st column, then read the load average from uptime and compare it with your core count from nproc.
If steal is near zero and your load average sits well below your core count, you have a sizing question, not a hosting question. A larger virtual plan is the cheaper answer.
If steal is consistently high, the host is oversubscribed. Move to a plan with better isolation, or to a different provider, before you conclude that virtualisation itself is the problem.
If steal is near zero and your load average is pinned at your core count around the clock, you have outgrown shared pricing. That is the honest signal to price dedicated hardware.
Run the same test after the move. A server that was slow because of a missing database index is still slow on bare metal, and it costs more there.
FAQ
Is a dedicated server always faster than a VPS?
No. A dedicated server gives you exclusive hardware, but a modern VPS on fast NVMe storage with a recent CPU can beat an older dedicated machine. Speed depends on the hardware generation, the storage, and how heavily the host is oversubscribed. Compare the specifications, then measure steal time on the VPS. A dedicated server wins reliably only when contention is the thing limiting you.
How do I know if a noisy neighbour is slowing my VPS?
Run vmstat 1 60 while the server is busy and read the st column, which is steal time: the percentage of time your virtual CPU was ready to run but the physical core was given to another guest. Near zero is healthy, a few percent under load is normal, and sustained double digits means the host is oversubscribed. High steal with low load inside your own server points at the host, not at your application.
Can I run a VPS and a dedicated server together?
Yes, and it is a common shape. Put the steady, resource hungry part on dedicated hardware, such as a database or a build farm, and keep the bursty parts on virtual plans you can resize or destroy freely, such as web front ends, staging and monitoring. Connect them over a private network or an encrypted tunnel and manage them as one fleet.
Is a VPS secure enough compared with dedicated hardware?
For almost every workload, yes. The hypervisor isolates memory and storage between guests, so another tenant cannot read your data. The real difference is contractual and regulatory: some audits require that no other tenant shares the physical machine, and only dedicated hardware meets that. Day to day, your patching, your firewall and your access control decide your security far more than the tenancy model does.
What happens when I outgrow my VPS?
Usually you resize it, which on most providers means picking a larger plan and rebooting. That path continues until you reach the largest plan offered, or until your load is saturated around the clock and the virtual plan costs more than equivalent hardware would. At that point, price a dedicated server and plan a migration, because the disks cannot follow you across the step.