What Is a VPS? A Plain-English Guide to Virtual Private Servers (2026)

A VPS (virtual private server) is a virtual machine you rent from a hosting provider: it has its own operating system, root access, virtual CPUs, memory, disk and public IP address, but it runs on a physical server shared with other customers. It behaves like a small Linux server you reach over SSH, while a hypervisor such as KVM keeps it isolated from the other virtual machines on the same hardware.

So the short answer to “what is a VPS” is: your own server, on someone else’s shared hardware. The rest of this guide covers what is shared, what fits on one, what it costs to run and what to do first.

Key takeaways

• A VPS is a virtual machine you rent from a provider: its own operating system, root access, RAM, disk and public IP, on a physical host shared with other virtual machines.
• A hypervisor such as KVM isolates the guests; what you share is the hardware, and on shared-vCPU plans the sharing shows up as steal time (the st column in top and vmstat).
• Size from documented minimums: Ubuntu Server 24.04 cloud images need 1 GB of RAM, while Claude Code and n8n's Docker Compose setup with its Assistant sandbox call for 4 GB.
• Short jobs suit hourly billing, and always-on servers need a monthly plan or an hourly bill capped at the monthly price; 12 of the 20 providers in the Hourly VPS Fine-Print Index keep billing a stopped server until it is deleted (checked October 2, 2026).
• Secure a new VPS before you expose anything: SSH keys, no root or password login, a firewall and updates.

VPS meaning: virtual, private, server

VPS stands for virtual private server. Each of the three words is doing real work:

• Virtual: the machine is defined in software. The provider’s hypervisor gives it processors, memory, a disk and a network card that look real to the operating system inside.
• Private: other customers on the same host cannot see your files, processes or memory, and the root account is yours alone. Private means isolated, not hidden: the provider knows who rents the server, and the server answers on a public IP address.
• Server: it is a computer that stays on and connected to the internet, so it keeps answering requests while your laptop is closed.

You will also meet other names for the same product: “virtual server”, “cloud server”, “instance” at the large clouds and “Droplet” at DigitalOcean. “VDS” (virtual dedicated server) is a label some sellers use for plans with dedicated CPU cores; it has no standard definition, so read the CPU line of the plan.

A VPC (virtual private cloud) is something else: a network, not a server. AWS defines its VPC as “a logically isolated virtual network that you’ve defined”, and servers attach to it.

Traders mean the same thing: a server that runs a trading platform or bot around the clock. The MetaTrader 5 help advises picking the hosting server closest to your broker to cut order latency. Outside computing, VPS is also the airport code for Destin–Fort Walton Beach, Florida, and a medical abbreviation for a ventriculoperitoneal shunt.

How does a VPS work?

A provider buys large physical servers, called hosts, and runs a hypervisor on each one. The hypervisor divides the host’s CPU time, memory, storage and network into virtual machines and keeps them apart. Each virtual machine boots its own kernel and operating system, exactly as a physical server would.

KVM (for Kernel-based Virtual Machine) is a full virtualization solution for Linux on x86 hardware containing virtualization extensions (Intel VT or AMD-V).

KVM project, linux-kvm.org

KVM turns the Linux kernel itself into the hypervisor, and it has been part of mainline Linux since version 2.6.20. The KVM project sums up the result: each virtual machine has “private virtualized hardware: a network card, disk, graphics adapter, etc.” Every HourlyVPS server is a KVM virtual machine, on hosts with Intel Xeon Platinum CPUs, DDR4 ECC registered memory and enterprise NVMe drives.

What you share with your neighbors is the physical box. What you get is a slice of each resource, and the four slices behave differently.

vCPU: shared or dedicated?

A vCPU is a virtual processor. On a KVM host each vCPU runs as an ordinary host thread; the Linux kernel’s KVM documentation calls it “the vCPU thread” and notes that it can move between logical processors and share one with other tasks. How those threads are scheduled decides what you are buying:

• Shared vCPU: your vCPUs take turns on a pool of physical cores with other customers’ vCPUs. You can use 100% when neighbors are quiet, but long stretches of heavy work compete for the same cores. Entry-level plans are usually shared: Akamai’s Nanode, Google’s e2-micro and Vultr’s Cloud Compute all are, per our Hourly VPS Fine-Print Index. That is why they cost little.
• Dedicated vCPU: each vCPU is typically pinned to host CPU threads reserved for your server, so other customers’ work does not run there. In libvirt, the common management layer for KVM, this is the vcpupin setting, which “specifies which of host’s physical CPUs the domain vCPU will be pinned to”. Pick it for builds, video encoding, load generators and busy game servers.

Linux measures the difference for you. When the hypervisor runs something else while your vCPU wants to work, your kernel counts it as steal time: “the time spent in other operating systems when running in a virtualized environment” (proc_stat(5)). The st column in top and vmstat shows it as a percentage. Our rule of thumb, not a standard: a few percent now and then is normal on shared vCPUs; a steady double-digit share during your busy hours means the job wants dedicated cores. To log steal time during a test and read it next to CPU and disk results, see the VPS benchmark guide.

At HourlyVPS, Quartz plans use shared vCPUs and Chrono plans use dedicated, pinned vCPUs.

RAM: the limit you hit first

The RAM in your plan is what your operating system sees as installed memory. The hypervisor does not time-share memory the way it time-shares CPU. When your applications need more than you have, Linux falls back to swap if you set it up, and when that runs out, its out-of-memory killer stops a process. Size RAM for your peak, not your average.

NVMe storage

Your disk is a virtual block device backed by drives in the host. NVMe (NVM Express), the command set for storage attached over PCI Express, matters most for databases, package installs and builds, which make many small reads and writes. Size the disk for the OS, apps, logs and container images. A snapshot is a rollback point before a risky change, not a backup.

Network, IPv4 and IPv6

The hypervisor gives your VPS a virtual network card connected to the host’s uplink, plus public IP addresses. Every server comes with one dedicated IPv4 address and IPv6, included in the plan price.

IPv4 matters because it is scarce. The RIPE NCC announced on November 25, 2019 that it had run out of IPv4 addresses, and 4 of the 20 providers in our Hourly VPS Fine-Print Index now bill IPv4 on top of the server price (checked October 2, 2026). IPv6 is no longer niche: on every day of September 2026, between about 46% and 51% of users reached Google over IPv6, according to Google’s IPv6 statistics.

Traffic is the other network number to read. Istanbul plans include the monthly traffic allowance listed for each plan, prorated for a server that exists for part of a billing period. Every HourlyVPS server also sits behind always-on inline DDoS mitigation with 1.2 Tbps of capacity; the network page has the details.

Is a VPS a virtual machine?

Yes. A KVM VPS is a virtual machine sold as a product: “VM” names the technology, “VPS” names what you rent.

The exception is the container-based “VPS” some budget hosts sell, built on OpenVZ or LXC. LXC’s own documentation places containers “in the middle between a chroot and a full fledged virtual machine”, running “without the need for a separate kernel”. You get root inside the container, but you share the host’s kernel, so you cannot choose your kernel version or load your own kernel modules, and software that needs kernel features, such as Docker, may be restricted. On a KVM VPS, Docker installs as on any Ubuntu or Debian server.

Option	What you rent	Own kernel?	Root access	Typical billing
Shared hosting	An account on a server shared with many sites	No	No	Monthly or yearly
Container “VPS” (OpenVZ, LXC)	An isolated container on a shared kernel	No	Inside the container	Monthly
KVM VPS	A full virtual machine with a fixed size	Yes	Yes	Monthly, daily or hourly
Cloud instance	A virtual machine plus separately metered disks, IPs and traffic	Yes	Yes	Per second, minute or hour
Dedicated server	The whole physical machine	Yes	Yes	Monthly

To find out what you are on, run systemd-detect-virt: it prints kvm on a KVM virtual machine and lxc or openvz inside those containers. For the full trade-offs against shared hosting and dedicated hardware, read VPS vs shared hosting vs dedicated vs cloud.

What is a VPS used for?

Anything that needs a Linux machine that is always reachable, or a clean machine you can throw away afterwards. The most common jobs, each with a step-by-step guide:

• Websites and APIs: install WordPress on a VPS, serve apps over HTTPS with a Caddy reverse proxy, package them as containers after you install Docker on the VPS, or deploy them from Git with Coolify, a self-hosted PaaS.
• Bots that must stay online: host a Discord bot or run a Telegram bot 24/7.
• Workflow automation: self-host n8n instead of paying per execution.
• Your own cloud services: a file cloud with Nextcloud on a VPS, or uptime checks and a status page with Uptime Kuma.
• AI agents and coding agents: give agents an always-on home (VPS for AI agents) or keep a remote dev box for Claude Code away from your laptop. Without a GPU, the VPS runs the agent and its tools while the model runs behind the provider’s API. Small open models also run on CPU, slowly; Ollama on a CPU VPS shows what fits in RAM.
• Game servers for a group: see what a Minecraft server costs or set up a Palworld dedicated server or a Valheim dedicated server.
• CI and testing: run a self-hosted GitHub Actions runner, or load test with k6 against systems you own or have written permission to test (see our acceptable use policy).
• Your own VPN endpoint: set up WireGuard on a VPS for a trip or a Tailscale exit node, within local law and the AUP.
• Learning: learn Linux on a VPS you can break, rebuild and delete.

A VPS is the wrong tool for GPU work such as training models or running large local language models, and for anyone who does not want to patch and secure an operating system.

Do you need a VPS?

Use this checklist. Find the row that matches your job and read across.

Your situation	Need a VPS?	Why, or what fits better
A blog or brochure site on a standard CMS, no custom software	Usually not yet	Shared or managed hosting is less work. Move up when you need root access or hit its limits.
A process that must run 24/7: bot, agent, worker, scheduled jobs	Yes	It needs a machine that is always on with a stable IP. Monthly billing, or hourly billing capped at the monthly price, fits.
You need root, custom ports, Docker or a specific OS version	Yes	Shared hosting usually offers none of these.
A clean Linux box for a few hours: test a config, open an untrusted repo, run a demo	Yes, hourly	Deploy, use, delete. You pay for the hours, and nothing touches your own machine.
A game server for weekend sessions	Yes, by the day or the hour	Run it while the group plays, delete it after.
Hours of 100% CPU: builds, encoding, load generation	Yes, with dedicated vCPU	On shared vCPUs this shows up as steal time.
Training models or running large local LLMs	No	You need GPU instances.
Hardware-level isolation or the whole machine’s resources	Probably not	Rent a dedicated server.
Only encrypting your browsing on public Wi-Fi	Not necessarily	A VPN service is simpler. A VPS lets you run your own WireGuard endpoint.
You do not want to maintain an operating system	No	A platform-as-a-service or a managed VPS does that for you; see the table below.

Managed or unmanaged VPS?

On an unmanaged VPS, the operating system and everything on it are yours to run. The split usually looks like this:

Task	Unmanaged VPS	Managed VPS
Hardware, hypervisor and network	Provider	Provider
OS updates and security patches	You	Usually the provider
SSH access, users and firewall	You, with full root	Usually the provider; root access may be limited
Backups	You	Often included; check what a restore covers
Your applications	You	Varies by plan, often through a control panel
Skills you need	Linux command line basics	Fewer

Unmanaged costs less and leaves every choice to you; managed costs more and suits people who would rather not run an operating system. HourlyVPS servers are unmanaged, as our security page states.

What size VPS do you need?

Start from the documented minimum of the software you run, add the operating system, then leave headroom for peaks. Canonical lists 1 GB of RAM and 4 GB of disk as the minimum for Ubuntu Server 24.04 LTS cloud images (1.5 GB and 5 GB for ISO installs), and suggests “3 GB or more” of RAM and “25 GB or more” of disk.

Workload	What the official docs say	Starting plan and reasoning	How long it runs
Learning Linux, one small service	Ubuntu Server 24.04 cloud images: 1 GB RAM, 4 GB disk minimum	Quartz Q1 (1 GB) meets the minimum; Q2 (2 GB) leaves room for upgrades and a small database	A few hours per session
One Discord or Telegram bot	No official minimum; one long-running process	Q1 or Q2: the OS minimum plus one small process (our reasoning)	24/7
n8n with Docker Compose	n8n’s sizing guidance: 320 MB to 2 GB of memory; the Docker Compose setup with the Assistant’s code sandbox needs at least 4 GB RAM and 2 vCPUs	Quartz Q4 (2 vCPU, 4 GB)	24/7
Claude Code remote dev box	Anthropic’s requirements: 4 GB+ RAM; Ubuntu 20.04+ or Debian 10+	Quartz Q4 (2 vCPU, 4 GB)	While you work, or 24/7
k6 load test with about 1,000 virtual users	k6 docs: about 1–5 MB of RAM per VU for simple tests (1,000 VUs = 1–5 GB); keep at least 20% of CPU idle	Chrono C8 (2 dedicated vCPU, 8 GB): room for the VUs and steady CPU	The test window, often a few hours
Palworld dedicated server	Pocketpair: 4 CPU cores recommended; 16 GB RAM, 32 GB or more recommended for stable operation; 8 GB boots but risks out-of-memory crashes	Chrono C16 (4 dedicated vCPU, 16 GB) for the 16 GB figure; Chrono C32 (8 dedicated vCPU, 32 GB) for the 32 GB recommendation	Play weekends, by the day
Small website: web server, PHP and a database	No single official RAM figure	Q2 to start, Q4 once the database grows (our reasoning)	24/7

Three common starting points, from a first server to steady CPU work:

Plan	vCPU, RAM, disk	Per hour	Monthly cap	Fits
Quartz Q1	1 shared vCPU, 1 GB RAM, 25 GB NVMe	$0.01/hour	$5.00/month cap	Covers the Ubuntu minimum: enough to learn on or to run one small bot
Quartz Q4	2 shared vCPU, 4 GB RAM, 80 GB NVMe	$0.03/hour	$15.00/month cap	Meets the 4 GB that Claude Code and n8n’s Docker Compose setup call for
Chrono C8	2 dedicated vCPU, 8 GB RAM, 100 GB NVMe	$0.07/hour	$35.00/month cap	Pins two vCPUs for jobs that keep the processor busy for hours

The middle option, Quartz Q4, in full, with its price per hour, per day and its monthly cap:

How does VPS billing work?

Traditional hosts sell a VPS by the month, sometimes with lower rates for multi-year terms. Cloud-style providers meter time instead: you deploy, the clock runs, you delete, the clock stops. The fine print differs a lot; for typical price levels, see how much a VPS costs.

Across the 20 providers in our Hourly VPS Fine-Print Index (checked October 2, 2026), 12 bill per hour, 3 per minute and 5 per second. Eleven cap hourly charges at a monthly price. Twelve keep billing a stopped server in full until it is deleted, typically because its resources stay reserved; the guide to which VPS providers bill stopped servers lists them. For three large clouds side by side, see Vultr vs DigitalOcean vs Hetzner billing.

Billing model	How you pay	Fits	Watch for
Monthly, prepaid	One payment per month	Servers that run all the time	Paying a full month for a short job; long contract terms
Hourly, metered	A rate for every hour the server exists	Tests, CI, demos, short projects, experiments	A forgotten server keeps billing; stopped servers often still bill
Hourly with a monthly cap	The hourly rate until the month’s charges reach the monthly price, then nothing more that month	Short jobs and always-on servers alike	When the cap starts (DigitalOcean and Vultr: after 672 hours on most plans, as of October 2026) and whether a stopped server still counts
Per second or minute	A rate per second or minute, often with a minimum charge	Automated jobs that live minutes	Minimum charges (DigitalOcean: 60 seconds or $0.01, whichever is higher, as of October 2026) and extras billed separately

How HourlyVPS bills a VPS

HourlyVPS bills every plan one way: by the hour, with a monthly cap, so there is no billing mode to choose up front. Charges are deducted in whole cents: a full hour costs exactly the hourly rate, and a partial first or last hour is rounded to the nearest cent, never more than a full hour. You pay the hourly rate for each hour a server exists until its charges reach the plan’s monthly price, after 500 hours (about 20.8 days), and the rest of that billing period (one month from your order date) is free. Your account runs on prepaid credit, with a minimum top-up of $5. How long you keep a server decides what it costs:

• Hourly VPS, for tests, demos and builds: deploy, use the server, delete it, and pay only for the hours it existed.
• Daily VPS, for weekends and events: a day is 24 hours of hourly billing, so three days are billed as 72 hours.
• Monthly VPS, for servers that stay on: the bill stops at the monthly cap each billing period, with no contract and no annual term.

Billing is by the hour: every hour a server exists is charged at the plan’s hourly rate. The price tapes and cost tables on this site count every started hour as a full hour, so they show the most a duration can cost; the cost calculator charges a partial hour to the nearest cent, as the bill does. Each new server starts with an initial credit, paid at checkout and used for that server’s hours. When its balance gets down to about a day of usage, a top-up invoice is issued, and a server whose balance reaches zero is suspended, then deleted 7 days later. Delete the server yourself and billing stops.

Quartz Q2: what each duration costs on hourly billing, with the monthly cap applied

Duration	Hours on the meter	Cost $0.02/hour · cap $10.00/month	Note
1 hour	1	$0.02
8 hours	8	$0.16
1 day	24	$0.48
2 days	48	$0.96
7 days	168	$3.36
30 days	720	$10.00	Capped at the monthly price

The practical upshot: you never have to guess how long a server will live. Delete it after an afternoon and you pay for the afternoon; leave it on and the bill stops growing at the monthly price for the rest of the billing period. The hourly vs monthly guide compares the cap with providers that bill every hour, the VPS cost calculator runs the numbers for your own hours, the hourly billing guide explains every rule, and the pricing page lists every plan.

What should you do first on a new VPS?

1. Pick the location closest to your users. Latency follows distance; how to choose a VPS location shows how to measure it. HourlyVPS runs in Istanbul today, with New York coming soon.
2. Choose an image such as Ubuntu 24.04 LTS or Debian. If the order form accepts an SSH public key, add it; otherwise you copy your key to the server in your first minutes.
3. Log in. Follow how to connect to a VPS over SSH from your own Windows, macOS or Linux computer.
4. Secure it before you install anything else. Update packages, create a sudo user, turn off root and password logins, and enable a firewall that opens only the ports you use. The checklist to secure a new Linux VPS has every command.
5. Check what you got with the commands below. If the server will send email, read the outbound port 25 rule first: SMTP on port 25 is closed on new servers and opens only on request, through a ticket.
6. Install your stack, for example Docker and a reverse proxy, then take a snapshot before risky changes.
7. When the job is done, delete the server. Stopping it does not stop billing. Work through the checklist before you delete a VPS, so no data or DNS record is left behind.

Once you are logged in as your sudo user, run these read-only commands at the shell prompt. They show what the hypervisor actually gave you:

systemd-detect-virt
nproc
free -h
lsblk
ip -brief address
vmstat 1 5

systemd-detect-virt prints kvm on a KVM virtual machine. nproc counts your vCPUs, free -h shows RAM and swap (if the Swap line reads 0B and you want a safety margin, see how to add swap on Ubuntu 24.04), lsblk lists the virtual disk, and ip -brief address lists your IPv4 and IPv6 addresses. In the vmstat output, the st column is the steal time described above.